Cogalois Theory

COGALOIS THEORY
PURE AND APPLIED MATHEMATICS

A Program of Monographs, Textbooks, and Lecture Notes
MONOGRAPHS AND TEXTBOOKS IN PURE AND APPLIED MATHEMATICS

1. K. Yano, lntegral Formulas in Riemannian Geometry (1970) 2. S. Kobayashi, Hyperbolic Manifolds and Holomorphic Mappings (1970) 3. V. S. Vladimimv, Equations of Mathematical Physics (A. Jeffrey, ed.; A. Lifflewood, trans.) (1970) 4. B. N. Pshenichnyi, Necessary Conditions for an Extremum (L. Neustadt, translation ed.; K. Makowski, trans.) (1971) 5. L. Narici et a/., Functional Analysis and Valuation Theory (1971) 6. S. S. Passman, Infinite Group Rings (1971) 7. L. Dornhoff, Group Representation Theory. Part A: Ordinary Representation Theory. Part B: Modular Representation Theory (1971,1972) 8. W. Boothby and G. L. Weiss, eds., Symmetric Spaces (1972) 9. Y. Matsushima, Differentiable Manifolds (E. T. Kobayashi, trans.) (1972) 10. L. E. Ward, Jr., Topology (1972) 11. A. Babakhanian, Cohomological Methods in Group Theory (1972) 12. R. Gilmer, Multiplicative Ideal Theory (1972) 13. J. Yeh, Stochastic Processes and the Wiener Integral (1973) 14. J. Barns-Neto, lntroductionto the Theory of Distributions (1973) 15. R. Larsen, Functional Analysis (1973) 16. K. Yano and S. Ishihara, Tangent and Cotangent Bundles (1973) 17. C. Pmcesi, Rings with Polynomial Identities (1973) 18. R. Hennann, Geometry, Physics, and Systems (1973) 19. N. R. Wallach, Harmonic Analysis on Homogeneous Spaces (1973) 20. J. Dieudonnd, lntroductionto the Theory of Formal Groups (1973) 21. 1 . Vaisman, Cohomology and Differential Forms (1973) 22. B.-Y. Chen, Geometry of Submanifolds (1973) 23. M. Marcus, Finite Dimensional Multilinear Algebra (in two parts) (1973, 1975) 24. R. Larsen, Banach Algebras (1973) 25. R. 0. Kujala and A. L. Viffer, eds., Value Distribution Theory: Part A; Part B: Deficit and Bezout Estimates by Wilhelm Stoll(1973) 26. K. 8. Stolarsky, Algebraic Numbers and DiophantineApproximation (1974) 27. A. R. Magid, The Separable Galois Theory of Commutative Rings (1974) 28. 8. R. McDonald, Finite Rings with Identity (1974) 29. J. Satake, Linear Algebra (S. Koh et at., trans.) (1975) 30. J. S. Golan, Localizationof Noncommutative Rings (1975) 31. G. Klambauer, Mathematical Analysis (1975) 32. M. K. Agoston, Algebraic Topology (1976) 33. K. R. Goodearl, Ring Theory (1976) 34. L. E. Mansfield, Linear Algebra with Geometric Applications (1976) 35. N. J. Pullman, Matrix Theory and Its Applications (1976) 36. B. R. McDonald, Geometric Algebra Over Local Rings (1 976) 37. C. W. Gmetsch, Generalized Inverses of Linear Operators (1977) . L. Gersting, Abstract Algebra (1977) 38. J. E. Kuczkowskiand J 39. C. 0. Christenson and W. L. Voxman, Aspects of Topology (1977) 40. M. Nagata, Field Theory (1977) 41. R. L. Long, Algebraic Number Theory (1977) 42. W. F. Pfeffer, Integrals and Measures (1977) 43. R. L. Wheeden and A. Zygmund, Measure and Integral (1977) 44. J. H. Curtiss, lntroduction to Functions of a Complex Variable (1978) 45. K. Hrbacek and T. Jech, lntroductionto Set Theory (1978) 46. W. S. Massey, Homology and Cohomology Theory (1978) 47. M. Marcus, lntroduction to Modem Algebra (1978) 48. E. C. Young, Vector and Tensor Analysis (1978) 49. S. 8.Nadler, Jr., Hyperspaces of Sets (1978) 50. S. K. Segal, Topics in Group Kings (1978) 51. A. C. M. van Rooij, Non-Archimedean Functional Analysis (1978) 52. L. Corwin and R. Szczarba, Calculus in Vector Spaces (1979) 53. C. Sadosky, Interpolationof Operators and Singular lntegrals (1979) 54. J. Cmnin, Differential Equations (1980) 55. C. W. Gmetsch, Elements of Applicable Functional Analysis (1980)
EXECUTIVE EDITORS
Earl J . Taft Rutgers University New Brunswick, New Jersey
Zuhair Nashed University of Delaware Newark, Delaware
EDITORIAL BOARD
M. S. Baouendi University of California, Sun Diego Jane Cronin Rutgers University Jack K. Hale Georgia Institute of Technology S. Kobayashi University of California, Berkeley Marvin Marcus University of California, Santa Barbara
W. S. Massey Yale University
Anil Nerode Cornell University Donald Passman University of Wisconsin, Madison Fred S. Roberts Rutgers University David L. Russell Virginia Polytechnic Institute and State University Walter Schempp Universitat Siegen Mark Teply University of Wisconsin, Milwaukee
1 . Vaisman, Foundations of Three-DimensionalEuclidean Geometry (1980) H. I. Freedan, DeterministicMathematical Models in Population Ecology (1980) S. B. Chae, Lebesgue Integration (1980) C. S. Rees et al., Theory and Applications of Fourier Analysis (1981) L. Nachbin, lntroduction to Functional Analysis (R. M. Aron, trans.) (1981) G. Onech and M. Orzech, Plane Algebraic Curves (1981) R. Johnsonbaugh and W. E. Pfaffenberger, Foundations of Mathematical Analysis (1981) 63. W. L. Voxman and R. H. Goetschel, Advanced Calculus (1981) 64. L. J. Corwin and R. H. Szczarba, Multivariable Calculus (1982) 65. V. I. IstrZjtescu, lntroduction to Linear Operator Theory (1981) 66. R. D. Jarvinen, Finite and Infinite Dimensional Linear Spaces (1981) 67. J. K. Beem and P. E. Ehrlich, Global Lorentzian Geometry (1981) 68. D. L. Armacost, The Structure of Locally Compact Abelian Groups (1981) 69. J. W. Brewer and M. K. Smith, eds., Emmy Noether: A Tribute (1981) 70. K. H. Kim, Boolean Matrix Theory and Applications (1982) 71. T. W. Wieting, The Mathematical Theory of Chromatic Plane Ornaments (1982) 72. D. B.Gauld, DifferentialTopology (1982) 73. R. L. Faber, Foundations of Euclidean and Non-Euclidean Geometry (1983) 74. M. Carmeli, Statistical Theory and Random Matrices (1983) 75. J. H. Carmth et a/., The Theory of Topological Semigroups (1983) 76. R. L. Faber, Differential Geometry and RelativityTheory (1983) 77. S. Bamett, Polynomials and Linear Control Systems (1983) 78. G. Karpilovsky, Commutative Group Algebras (1983) 79. F. Van Oystaeyen and A. Verschoren, Relative Invariants of Rings (1983) 80. 1 . Vaisman, A First Course in Differential Geometry (1984) 81. G. W. Swan, Applications of Optimal Control Theory in Biomedicine (1984) 82. T. Petrie and J. D. Randall, Transformation Groups on Manifolds (1984) 83. K. Goebel and S. Reich, Uniform Convexity. Hyperbolic Geometry, and Nonexpansive Mappings (1984) 84. T. Albu and C. Ndstdsescu, Relative Finiteness in Module Theory (1984) 85. K. Hrbacek and T. Jech. lntroduction to Set Theory: Second Edition (1984) 86. F. Van Oystaeyen and A. Verschoren, Relative lnvariants of Rings (1984) 87. B. R. McDonald, Linear Algebra Over Commutative Rings (1984) 88. M. Namba, Geometry of Projective Algebraic Curves (1984) 89. G. F. Webb, Theory of Nonlinear Age-Dependent Population Dynamics (1985) 90. M. R. Bremner et a/., Tables of Dominant Weight Multiplicities for Representations of Simple Lie Algebras (1985) 91. A. E. Fekete, Real Linear Algebra (1985) 92. S. B. Chae, Holomorphy and Calculus in Normed Spaces (1985) 93. A. J. Jeni, lntroduction to Integral Equations with Applications (1985) 94. G. Karpilovsky, Projective Representationsof Finite Groups (1985) 95. L. Narici and E. Beckenstein, Topological Vector Spaces (1985) 96. J. Weeks, The Shape of Space (1985) 97. P. R. Gribik and K. 0. Kortanek, Extremal Methods of Operations Research (1985) 98. J.-A. Chao and W. A, Woyczynski, eds., Probability Theory and Harmonic Analysis (1986) 99. G. D. Crown et a/., Abstract Algebra (1986) 100. J. H. Carmth et a/., The Theory of Topological Sernigroups, Volume 2 (1986) 101. R. S. Doran and V. A. Belfi, Characterizationsof C*-Algebras (1986) 102. M. W. Jeter, Mathematical Programming (1986) 103. M. Altman, A Unified Theory of Nonlinear Operator and Evolution Equations with Applications (1986) 104. A. Verschoren, Relative lnvariants of Sheaves (1987) 105. R. A. Usmani, Applied Linear Algebra (1987) 106. P. Blass and J. Lang, Zariski Surfaces and Differential Equations in Characteristic p > 0 (1987) 107. J. A. Re~ieke et a/., Structured Hereditary Systems (1987) 108. H. Busemann and 8.B. Phadke, Spaces with Distinguished Geodesics (1987) 109. R. Harte, lnvertibilityand Singularity for Bounded Linear Operators (1988) 110. G. S. Ladde et a/., Oscillation Theory of Differential Equations with Deviating Arguments (1987) 111. L. Dudkin et al., lterative Aggregation Theory (1987) 112. T. Okubo, Differential Geometry (1987)
56. 57. 58. 59. 60. 61. 62.
113. 114. 115. 116. 117. 118. 119. 120. 121. 122. 123. 124. 125. 126. 127. 128. 129. 130. 131. 132. 133. 134. 135. 136. 137. 138. 139. 140. 141. 142. 143. 144. 145. 146. 147. 148. 149. 150. 151. 152. 153. 154. 155. 156. 157. 158. 159. 160. 161. 162. 163. 164. 165. 166. 167. 168. 169. 170. 171.
D. L. Stancl and M. L. Stancl, Real Analysis with Point-Set Topology (1987) T. C. Gard, lntroduction to Stochastic Differential Equations (1988) S. S. Abhyankar, Enumerative Combinatorics of Young Tableaux (1988) H. Strade and R. Famsteiner, Modular Lie Algebras and Their Representations(1988) J. A. Huckaba, Commutative Rings with Zero Divisors (1988) W. D. Wallis, Combinatorial Designs (1988) W. Wieslaw, Topological Fields (1988) G. Karpilovsky, Field Theory (1988) S. Caenepeel and F. Van Oystaeyen, Brauer Groups and the Cohomology of Graded Rings (1989) W. Kozlowski, Modular Function Spaces (1988) E. Lowen-Colebunders, Function Classes of Cauchy Continuous Maps (1989) M. Pavel, Fundamentals of Pattern Recognition (1989) V. Lakshmikantham et al., Stability Analysis of Nonlinear Systems (1989) R. Sivaramakrishnan, The Classical Theory of Arithmetic Functions (1989) N. A. Watson, Parabolic Equations on an lnfinite Strip (1989) K. J. Hastings, lntroduction to the Mathematics of Operations Research (1989) B. Fine, Algebraic Theory of the Bianchi Groups (1989) D. N. Dikranjan et a/., Topological Groups (1989) J . C. Morgan 1 1 , Point Set Theory (1990) P. Biler and A. Witkowski, Problems in Mathematical Analysis (1990) H. J. Sussmann, Nonlinear Controllability and Optimal Control (1990) J.-P. Florens et a/., Elements of Bayesian Statistics (1990) N. Shell, Topological Fields and Near Valuations (1990) B. F. Doolin and C. F. Martin, lntroduction to Differential Geometry for Engineers (1990) S. S. Holland, Jr., Applied Analysis by the Hilbert Space Method (1990) J. Okninski, Semigroup Algebras (1990) K. Zhu, Operator Theory in Function Spaces (1990) G. 8. Price, An lntroduction to Multicomplex Spaces and Functions (1991) R. B. Darst, lntroduction to Linear Programming (1991) P. L. Sachdev, Nonlinear Ordinary Differential Equations and Their Applications (1991) T. Husain, Orthogonal Schauder Bases (1991) J. Foran, Fundamentals of Real Analysis (1991) W. C. Brown, Matrices and Vector Spaces (1991) M. M. Rao and Z. D. Ren, Theory of Orlicz Spaces (1991) J. S. Golan and T. Head, Modules and the Structures of Rings (1991) C. Small, Arithmetic of Finite Fields (1991) K Yang, Complex Algebraic Geometry (1991) D. G. Hoffman et a/., Coding Theory (1991) M. 0. Gonzalez, Classical Complex Analysis (1992) M. 0. Gonzalez, Complex Analysis (1992) L. W. Baggett, Functional Analysis (1992) M. Sniedovich, Dynamic Programming (1992) R. P. Agarwal, Difference Equations and Inequalities (1992) C. Brezinski, Biorthogonality and Its Applications to Numerical Analysis (1992) C. Swartz, An lntroduction to Functional Analysis (1992) S. B. Nadler, Jr., Continuum Theory (1992) M. A. Al-Gwaiz, Theory of Distributions (1992) E. Peny, Geometry: Axiomatic Developmentswith Problem Solving (1992) E. Castillo and M. R. Ruiz-Cobo, Functional Equations and Modelling in Science and Engineering (1992) A. J. Jem, Integral and Discrete Transforms with Applications and Error Analysis (1992) A. Charlier et al. , Tensors and the Clifford Algebra (1992) P. Biler and T. Nadzieja, Problems and Examples in Differential Equations (1992) E. Hansen, Global Optimization Using Interval Analysis (1992) S. Guerre-Delabriere, Classical Sequences in Banach Spaces (1992) Y . C. Wong, Introductory Theory of Topological Vector Spaces (1992) S. H. Kulkami and B. V. Limaye, Real Function Algebras (1992) W. C. Brown, Matrices Over Commutative Rings (1993) J. Loustau and M. Dillon, Linear Geometry with Computer Graphics (1993) W. V. Petryshyn, Approximation-Solvability of Nonlinear Functional and Differential Equations (1993)
COGALOIS THEORY
TOMA ALBU
Atilim University Ankara, Turkey, and Bucharest University Bucharest, Romania
MARCEL
MARCEL DEKKER, INC.

DEKKER
NEWYORK BASEL
Library of Congress Cataloging-in-PublicationData A catalog record for this book is available from the Library of Congress. ISBN: 0-8247-0949-7
This book is printed on acid-free paper.
To my beloved wife Marina
Headquarters Marcel Dekker, Inc. 270 Madison Avenue, New York. NY 10016 tel: 212-696-9000; fax: 2 12-685-4540 Eastern Hemisphere Distribution Marcel Dekker AG Hutgasse 4, Postfach 8 12, CH-400 1 Basel, Switzerland tel: 4 1-61-260-6300; fax: 4 1-61-260-6333 World Wide Web http://www.dekker.com
The publisher offers discounts on this book when ordered in bulk quantities. For more information, write to Special Sales/Professional Marketing at the headquarters address above.
Copyright O 2003 by Marcel Dekker, Inc. All Rights Reserved.

Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, microfilming, and recording, or by any information storage and retrieval system, without permission in writing from the publisher. Current printing (last digit): 1 0 9 8 7 6 5 4 3 2 1
PRINTED IN THE UNITED STATES OF AMERICA
Preface
interesting but difficult problem in Field Theory is t o describe in a satisfactory manner the lattice Intermedkte ( E I F ) of all intermediate fields of a given field extension E I F . If EIF is a finite Galois extension, then, by the Fundamental Theorem of Galois Theory, the lattice Intermediate ( E I F ) is anti-isomorphic t o the lattice Subgroups(Gal(E/F)) of all sllbgroups of the Galois group G a l ( E / F ) of the extension E I F . On t,he other hand, there exists a fairly large class of field extensions t h a t are not necessarily Galois, but which enjoy a property that is dual t o the previous one. Namely, these are the extensions EIF for which there exists a. canonical lattice isomorphism (and not a lattice anti-isomorphism as in the Galois case) between Intermediate ( E I F ) and Subgroups ( A ) , where A is a certain group canonically associated with the extension E I F . \Ye call the members of this class, extensions with A-Cogalois correspondence. Their prototype is the field extension Q , . . . "i/ar)/Q, is where r, n l , . . . , n,, a l , . . . , a , are positive integers, and where &n the positive real ni-th root of ai for each i , 1 5 i 5 r. For such an extension, the associated group A is the factor group @ ("*, . . . , "&)I Q .
A11
("m .
The purpose of this monograph is t o provide a systematic investiga.tkm of field extensions which possess a Cogalois correspondence. This topic, we narned Cogalois Theory, is dual t o the very classical one known as Galois Theory investigating field extensions possessing a Galois correspondence. -4s is well-known, Galois Theory has a finite part dealing with finite Galois extensions, as well as an infinite part dealing with infinite Galois extensions. Similarly, Cogalois Theory has a. finite (resp. infinite) part,, studying finite (resp. infinite) field extensions with a Cogalois correspondence. The first part of this monograph is devoted t o the study of finite field extensions with a Cogalois correspondence. T h e case of infinite extensions is discussed in the second part of the monograph.
Preface
Preface
vii
An effort was made t o keep the account as self-contained as possible. However, a certain knowledge of Field Theory and Group Theory, including the Fundamental Theorem of Galois Theory, on the level of a gradua,te course is needed for a good understanding of the most part of this volume. Thus, we assume from the reader some familiarity with the notions and basic facts on vector spaces, groups, rings, Galois extensions, as presented e.g., in the books of Bourbaki [40], Kaplansky [73], Karpilovsky [76], or Lang [80], although, whenever it was possible, the indispensable basic facts on these notions were included in the text. The reader is also assumed to have familiarity with some Galois cohomology, including crossed homornorphisrns, 1-coboundaries, the first group of cohomology, Hilbert's Theorem 90, etc., as presented e.g., in the books of Cassels and Frohlich [46], Karpilovsky [76], or Serre [96]. Some basic facts on topological groups, projective limits, and Pontryagin duality for discrete or compact -4belian groups are needed only in Chapters 14 and 15, for which the reader can be referred t o Hewitt and Ross [72], Karpilovsky [76], or Pontryagin [87]. Basics of Algebraic Kumber Theory are needed only in Section 9.3. They include algebraic number fields, algebraic integers, ideal class groups, Dedekind domains, class fields, etc., and may be found e.g., in the books of Hasse [67], Hecke [71], Keukirch [853, or Ribenboirn [89]. Elements of graded rings and of Hopf algebras are required only in Chapter 10, and may be found e.g., in the books of Caenepeel [42], D5sc5lescu, Niistiisescu, and Raianu [54], Kiistkescu and Van Oystaeyen [84], and Sweedler [102].
He also gratefully acknowledges partial financial support from grant D-7 awarded by the CNCSIS, Romania. For A . S - HT@ assistance during the preparation of this monograph, the author is very indebted t o Giinter Krause and Robert Wisbauer. The author would like t o thank Serban Basarab, Eberhard Becker, Sorin Diisclilescu, Laurenfiu Panaitopol, Serban Raianu, Marcel Tena, and Robert Wisbauer for helpful discussions on various subjects of this monograph. The author would also like to thank the anonymous referee for careful reading of the manuscript and valuable comments and suggestions which improved the presentation of the book, and especially for providing a completely different approach t o the main result of Chapter 15 (Theorem 15.3.8), which is sketched a t the end of that chapter. Last, but not least, the author is very indebted to Ralph Svetic for his great stylistic help.
ACKNOWLEDGEMENTS
This work was started by the author during his stay a t the Lniversity of California a t Santa Barbara as a Visiting Professor in the Spring Quarter of 1999. A large part of this work was done during his stay as a Humboldt Fellow at the University of Dortmund and Heinrich-Heine-University of Diisseldorf in July 2000 and April-June 2001. It was completed during his stay at the Atilim Cniversity, Ankara in the academic year 2001-2002. The author thanks all these institutions for their hospitality and financial support.
Contents
PREFACE INTRODCCTIOK
Part 1.
FINITE COGALOIS THEORY
Chapter 1. PRELIMINARIES 1.1. General notation and terminology 1.2. A short review of basic Field Theory 1.3. The Vahlen-Capelli Criterion 1.4. Bounded Abelian groups 1.5. Exercises t o Chapter 1 1.6. Bibliographical comments to Chapter 1 Chapter 2. KNESER EXTENSIONS 2.1. G-Radical and G-Kneser extensions 2.2. The Kneser Criterion 2.3. Exercises to Chapter 2 2.4. Bibliographical comments t o Chapter 2 Chapter 3. COGALOIS EXTEIVSIONS 3.1. The Greither-Harrison Criterion 3.2. Examples and properties of Cogalois extensions 3.3. The Cogalois group of a quadratic extension 3.4. Exercises to Chapter 3 3.5. Bibliographical cominents to Chapter 3 Chapter 4. STRONGLY KNESER EXTENSIONS 4.1. Galois and Cogalois connections 4.2. Strongly G-Kneser extensions 4.3. G-Coga.lois extensions
Contents
4.4. The Kneser group of a G-Cogalois extension 4.5. Almost G-Cogalois extensions 4.6. Exercises to Chapter 4 4.7. Bibliographical comments to Chapter 4 Chapter 5. GALOIS G-COGALOIS EXTENSIONS 5.1. Galois G-radical extensions 5.2. Abelian G-Cogalois extensions 5.3. Applications t o elementary Field Arithmetic (I) 5.4. Exercises to Chapter 5 5.5. Bibliographical comments t o Chapter 5 Chapter 6. RADICAL EXTENSIOKS AND CROSSED HOMOMORPHISMS 6.1. Galois extensions and crossed homomorphisms 6.2. Rgdical extensions via crossed homomorphisms 6.3. Exercises to C h p t e r 6 6.4. Bibliographical comments t o Chapter 6 Chapter 7. EXAMPLES OF G-COGALOIS EXTENSIONS 7.1. Classical Kumrner extensions 7.2. Generalized Kummer extensions 7.3. Kummer extensions with few roots of unity 7.4. Quasi-Kummer extensions 7.5. Cogalois extensions 7.6. Exercises t o Chapter 7 7.7. Bibliographical comments t o Chapter 7 Chapter 8. G-COGALOIS EXTENSIONS AND PRIMITIVE ELEMENTS 8.1. Primitive elements for G-Cogalois extensions 8.2. -4pplications to elementary Field Arithmetic (11) 8.3. Exercises to Chapter 8 8.4. Bibliographical comments t o Chapter 8 Chapter 9. APPLICATIONS T O ALGEBRAIC NUMBER. FIELDS 9.1. Number theoretic preliminaries 9.2. Some classical results via Cogalois Theory 9.3. Hecke systems of ideal numbers
9.4. Exercises to Chapter 9 9.5. Bibliographical cornments to Chapter 9 Chapter 10. COh'KECTIONS WITH GRADED ALGEBRAS -4ND HOPF ALGEBRAS 10.1. G-Cogalois extensions via strongly graded fields 10.2. Cogalois extensions and Hopf algebras 10.3. Exercises t o Chapter 10 10.4. Bibliographical comments t o Chapter 10
Part 2.
Chapter 11.1. 11.2. 11.3. 11.4. Chapter 12.1. 12.2. 12.3. 12.4. Chapter 13.1. 13.2. 13.3. 13.4. 13.5. 13.6.
INFINITE COGALOIS THEORY

11. INFINITE KNESER EXTENSIONS Infinite G-Kneser extensions Infinite strongly Kneser extensions Exercises to Chapter 11 Bibliographical comments to Chapter 11 12. INFINITE G-COGALOIS EXTENSIONS The General Purity Criterion and its applications Infinite Cogalois extensions Exercises to Chapter 12 Bibliographical comments to Chapter 12 13. IIC'FIKITE KUMMER THEORY Infinite classical Kurnmer extensions Infinite generalized Kummer extensions Infinite Kumrner extensions with few roots of unity Infinite quasi-Kummer extensions Exercises to Chapter 13 Bibliographical comments t o Chapter 13
Chapter 14. INFIKITE GALOIS THEORY AND PONTRYAGIN DUALITY 14.1. Profinite groups and Infinite Galois Theory 14.2. Character group and Pontryagin Duality 14.3. Exercises to Chapter 14 14.4. Bibliographical comments to Chapter 14
Chapter 15. 15.1. 15.2. 15.3. 15.4. 15.5.
INFINITE GALOIS G-COGALOIS EXTENSIONS The infinite Kneser group via crossed homomorphisms Lattice-isomorphic groups Infinite Abelian G-Cogalois extensions Exercises to Chapter 15 Bibliographical comments to Chapter 15
Bibliography Index
COGALOIS THEORY
INTRODUCTION
Consider the following very general problem in Mathematics: Describe i n a satisfactory manner the collection of all subobjects of a given object of a category C. In general, this is a difficult problem, but sometimes it can be reduced t o describing the subobjects of an object in another more suitable category V. For instance, let F be a given field and let C denote the category of all field extensions of F . Let E be an of object of C, i.e., a field extension E I F , and denote by E the set of all subfields of E containing F , in other words, the set of all subobjects of E in C. This set is, in general, a complicated-to-conceive, potentially infinite set of hard-to-describe-andidentify objects. However, when E/F is a finite Galois extension, then by the Fundamental Theorem of Galois Theory, there exists a canonical one-to-one order-reversing correspondence, or equivalently, a lattice antiisomorphism between the lattice & and the lattice of all subgroups of a certain group r, namely the Galois group Gal ( E I F ) , canonically associated with the extension E/F; we can say that such an EIF is an extension with r-Galois correspondence. In this way, the lattice of all subobjects of an object E E C, which has the additional property that it is a finite Galois extension of F, can be described by the lattice of all subobjects of the object I? in the category of all finite groups; in principle, this category is more suitable than the category C of all field extensions of F, since the set of all subgroups of a finite group is a far more benign object. Thus, many questions concerning a field are best studied by transforming them into group theoretical questions in the group of automorphisms of the field. So, the following natural problem arises:
Problem 1. Find large classes 3 of finite field extensions for which the lattice of all intermediate fields of every E/F E .F can be described in terms of the lattice of all subgroups of a certain group canonically associated with E I F .
Iritroduction
As we have already mentioned, finite Galois extensions are extensions as desired in Problem 1. On the other hand, there exist large classes of finite field extensions E/F which are not necessarily Galois, but are radical extensions of a special type, for which there exists a canonical one-toone order-preserving correspondence, or equivalently, a lattice isomorphism (and not a lattice anti-isomorphism as in the Galois case) between the lattice & of all intermediate fields of the extension E/F and the lattice V of all subgroups of a certain group A canonically associated with the extension E / F . It seems therefore very natural to introduce the following definition: a field extension E/F is said t o be an extension with A-Cogalois correspondence if there exists a canonical lattice isomorphism between the lattices & and V defined above. The term extension with Cogalois correspondence was introduced by Albu and Nicolae in [I91 to emphasize a situation which is, as shown above, somewhat dual t o that appearing in Galois Theory. To the best of our knowledge, the term of "Cogalois" appeared for the first time in the literature in 1986 in the fundamental paper of Greither and Harrison [63], where Cogalois extensions were introduced and investigated. Their pioneering work on Cogalois extensions was continued in 1991 by Barrera-Mora, n;l,eclowski-Calderh, and Villa-Salvador [30]. The prototype of extension with a A-Cogalois correspondence is any field extension where r E N* = N \ {0), n l , . . . , n,, a l , . . . ,a, E N*, and for every i , 1 6 i 6 r, is the positive real ni-th root of ai. For such an extension, the associated group A is the factor group @ ("-, ... , @. It seems surprising it was stated and proved explicitly only fairly late, in 1986, by Greither and Harrison [63] that such extensions are extensions with A-Cogalois correspondence. In particular, it follows that
As an immediate consequence of this equality one deduces that for any n E N* . Another proof of the last equality, using elementary Galois Theory, was given by Richards [go] (see also Gall's book [58], where the proof of Richards is reproduced). A result of the same nature was established by Lrsell [106]. A generalization of Besicovitch's result t o algebraic number fields satisfying certain conditions was established by Mordell [83]. Mordell's Theorem was further extended by Siege1 [98] and Schinzel [91]. All these results deal with a particular case of the following:
Problem 2. Let F be a field, let F be a n algebraic closure of F, and let X I , .. . , x, E p be elements of degree nl ,. . . ,n, over F, respectively. W h e n does the field F ( x l , . . . ,x,) have the degree nl . . . . - n, over F ?
A more general problem is the following:
Problem 3. W i t h the s a m e notation and hypotheses as in Problem 2, when can w e find a n explicit formula t o compute [ F ( x l , . . . ,a,) : F]?
Partial answers to Problem 3 are given by a well-known result on classical finite Kurnmer extensions (see, e.g., Artin [26]), as well as by a result appearing in Kaplansky's book [74, Theorems 601, and by another one of similar nature due t o Baker and Stark [28]. 111 his paper [77], Kneser answered Problem 3 for a large class of extensions that were called by Albu and Nicolae [19] K n e s e r extensions, honoring his crucial result. It is well-known that for any finitely many elements X I , .. . ,x, in F which are separable over an infinite field F, there exist elements cl, . . . ,c, in F such that B = CIGiG, cixi is a primitive element of the finite separable extension F ( x l , . . . ,x, ) / F , i.e., F (xl , . . . ,x,) = F(6). The following natural problem arises:
This equality was established for the first time in 1940, but only in the following particular case, by Besicovitch [35]: Let r E N* , and let pl, . . . , p , be different positive prime integers. If b l , . . . , b, are positive integers not divisible by any of these primes, a1 = b l p l , . . . ,a, = b,p,, and n l , . . . ,n, are arbitrary positive integers, then [Q('fi,.
Problem 4. Let F be a n y field, and let X I , . . . , x, in F be finitely m a n y xi a primitive element of the separable elements over F . W h e n i s finite separable extension F ( x l , . . . ,x,)/F ?
xlbi6,
.. ," ) f i
: Q ] = nl . . . . . n,.
Partial answers to this problem are given e.g., by Albu [3],Kaplansky [74, Theorem 641, and Zhou [113]. These were extended by Albu and Nicolae in [20], where a general statement was proved for the large class of finite
Introduction
Introduction
separable field extensions with A-Cogalois correspondence. In particular, from this general approach it follows very easily that where r E N* and n l , . . . , n , , a l , . . . ,a, E N*. An immediate consequence of this equality is the following result, which is folklore:
of infinite G-Cogalois extensions is performed in the second part, which contains 5 chapters. Now we will describe the contents of the first part of this volume. Chapter 1 contains the necessary preliminaries, including the terminology and notation which will be used throughout the book, a short review of basic Field Theory, the Vahlen-Capelli Criterion dealing with the irreducibility over an arbitrary field of binomials Xn - a , and some general facts on bounded Abelian groups. Chapter 2 introduces and investigates two basic concepts of Cogalois Theory, that of the G-radical extension and that of the G-Kneser extension. Roughly speaking, a radical extension is a field extension E/F such that E is obtained by adjoining t o the base field F an arbitrary set of "radicals" over F , i.e., of elements x E E such that xT1= a E F for some n E W . Such an x is denoted by fi and is called an n-th radical of a. So, E/F is a radical extension when E = F ( R ) , where R is a set of radicals over F . Clearly, one can replace R by the subgroup G = F*( R ) of the multiplicative group E* of E generated by F* and R. Thus, any radical extension E/F has the form E = F ( G ) , where G is a subgroup of E* containing F', with G / F * a torsion group. Such an extension is called Gradical. A finite extension E/F is said to be G-Kneser when it is G-radical and IG/F*I = [E : F ] . These extensions were introduced by Albu and Kicolae in [19], honoring the nice criterion due t o Kneser [77] evaluating the degrees of separable finite radical extensions. A complete proof of this criterion, which is a basic tool in this monograph, is provided in this chapter. Chapter 3 studies Cogalois extensions, which were introduced by Greither and Harrison in [63]. Using the concept of Kneser extension, a Cogalois extension is nothing else than a field extension E/F which is T ( E / F ) Kneser, where T ( E / F ) is the subgroup of the multiplicative group E* of the field E such that the factor group T ( E / F ) / F * is the torsion group of the factor group E * / F * . The group T ( E / F ) / F * , called by Greither and Harrison the Cogalois group of the extension E / F , is denoted by Cog(E/ F ) . Using the Kneser Criterion we provide a very simple and short proof of the Greither-Harrison Criterion [63] characterizing Cogalois extensions. Let us mention that the original proof in [63] of this criterion involves cohomology of groups. Then, we calculate effectively the Cogalois group of quadratic extensions of Q.
considered above, Greither and Harrison produced in [63] two other large classes of field extensions with A-Cogalois correspondence, namely the Cogalois extensions and the neat presentations. As is known, the finite classical Kummer extensions have a privileged position: they are at the same time extensions with Galois and with Cogalois correspondences, and the two groups appearing in this setting are isomorphic. Thus, the extensions of type Q . . . , "&)/Q, the Cogalois extensions, the neat presentations, and the finite classical Kummer extensions all are large classes of finite field extensions answering Problem 1.
"fi+...+ " G E Q M "&EQ Besides the extension Q ("C/aT, . . . , "*)/Q
forall l < i < r .
("m,
-4systematic investigation of separable finite radical extensions with 4-Cogalois correspondence was initiated by Albu and Kicolae in [19], and continued in [20], [21], [22], and [23]. They introduced the general concept of G- Cogalois extension, which permitted a very simple, unifying, and even more general approach t o the apparently different four classes of radical field extensions with a Cogalois correspondence presented above. These G-Cogalois extensions have many nice properties which answer in positive all the problems stated above.
The purpose of this monograph is to provide a systematic investigation of field extensions which possess a Cogalois correspondence. This subject can be named Cogalois Theory, and is dual t o the very classical one known as Galois Theory investigating field extensions possessing a Galois correspondence. As is well-known, Galois Theory has a finite part dealing with finite Galois extensions, as well as an infinite part dealing with infinite Galois extensions. Similarly, Cogalois Theory has a finite (resp. infinite) part, studying finite (re'sp. infinite) field extensions with a Cogalois correspondence. The investigation of finite G-Cogalois extensions is the body of the first part, comprising 10 chapters, of this monograph, while the investigation
Introduction
Chapter 4 contains the main results of this monograph. After presenting a very general discussion of Galois connections and Cogalois connections, we associate with any G-radical extension E I F , finite or not, a canonical Cogalois connection:
EIF is simultaneously G-Cogalois and H-Cogalois, then necessarily G = H. Consequently, it makes sense to define the Kneser group of a G-Cogalois
extension as the group G I F * , which will be denoted by Kne ( E I F ) . Chapter 4 ends with a study of almost G-Cogalois extensions, introduced by Lam-Estrada, Barrera-Mora, and Villa-Salvador [79] under the name of pseudo G- Cogalois extensions. Chapter 5 investigates field extensions which are simultaneously Galois and G-Cogalois. It is shown that whenever E/F is an Abelian GCogalois extension, then the Kneser group Kne (E/F ) and the Galois group Gal ( E I F ) of EIF are isomorphic in a noncanonical way. Applications of this result t o the elementary arithmetic of fields are also given.
&
where
1CI
cP
__e
G,
v:E--,G, $:G--+&>
E is the lattice Intermediate ( E / F ) of all intermediate fields of the extension E I F , and G is the lattice { H I F* 6 H G ) . Then, the basic notion of strongly G-Kneser extension is introduced: a finite field extension E/F is said to be strongly G-Kneser if it is a G-radical extension such that for any subfield K of E containing F the extension K / F is K * nG-Kneser, or equivalently, the extension E/ K is K *G-Kneser. These are precisely the G-Kneser extensions for which the maps
<
In the next chapter we investigate finite Galois extensions which are radical, Kneser, or G-Cogalois, in terms of crossed homomorphisms. The basic result of this chapter is description of the Cogalois group Cog(E/F) of any finite Galois extension E/F by means of crossed homomorphisms of the Galois group Gal(E/F) with coefficients in the group p ( E ) of all roots of unity in E . Chapter 7 provides three main classes of G-Cogalois extensions, namely the generalized Kummer extensions including the finite Kummer extensions with few roots of unity as well as the finite classical Kummer extensions, the Cogalois extensions, and the quasi-Kummer extensions. In particular, a series of results by Albu [3] and Greither and Harrison [63], as well as the essential part of the classical Finite Kummer Theory are re-obtained in a unified manner, more easily, and in a more general setting. Kote that the neat presentations introduced and investigated by Greither and Harrison [63] are very particular cases of quasi-Kumrner extensions, namely they are precisely those quasi-Kummer extensions which are Galois extensions of fields of characteristic 0. The prototype of finite Kummer extensions with few roots of unity is any subextension of R/Q of the form Q ( " n , .. . , 'G)/Q, with nl , . . . ,n,, a l , . . . ,a, positive integers. Notice that these extensions, in general, are not Galois. In particular, we derive from our Cogalois Theory results on finite Kummer extensions with few roots of unity, which are very similar to the known ones for classical finite Kummer extensions. Chapter 8 gives a complete answer t o Problem 4 within the class of G-Cogalois extensions. The main result of this chapter is the following. Let EIF be a G-Cogalois extension, let n E W ,and let (xi)lGi<TLbe
a : Intermediate ( E I F ) + Subgroups ( G I F * ) , a ( K ) = (K n G ) / F * ,
and
p : Subgroups ( G I F * ) + Intermediate ( E I F ) , P ( H / F * ) = F ( H ) ,
are isomorphisms of lattices, inverse to one another; in other words, the G-Kneser extensions E/F with G/F*-Cogalois correspondence. In the theory of field extensions with G/F*-Cogalois correspondence the most interesting are those which additionally are separable. They were called G-Cogalois extensions by Albu and Nicolae [19], and are completely characterized within the class of G-Kneser extensions by means of a very useful n-Purity Criterion, where n is the exponent of the finite group G I F * . This allows to obtain in a simple and unified manner, and even in a more general setting, a series of results from the classical Kummer Theory, as well as results of Albu [3]concerning Kurnmer extensions with few roots of unity, of Barrera-Mora, R~edowski-Calderbn,and Villa-Salvador [30], and of Greither and Harrison [63] concerning Cogalois extensions and neat presentations. It is shown that a separable G-Kneser extension EIF is G-Cogalois if and only if the group G I F * has a prescribed structure. As a consequence, the uniqueness of the group G is deduced. This means that if the extension
Introduction
a finite family of elements of G such that x i F * # x j F * for every i, j E { I , . . . , n ) , i # j. Then xl + . . . + x, is a primitive element of E/F if and only if G = K*(xl , . . . ,x,). A series of applications t o the elementary arithmetic of fields are also given. The Kneser Criterion has nice applications not only in investigating field extensions with Cogalois correspondence, but also in proving some results in Algebraic Number Theory. Chapter 9 presents such applications. It starts with a section devoted t o presenting basic nota,tion, terminology, and facts on algebraic number fields needed in the next sections of the chapter. A series of classical theorems due t o Hasse 1681, Besicovitch [35], Mordell [83],and Siege1 1981 related to solving Problem 2, as well as the links between them are presented in the second section of this chapter. The next section deals with a surprising application of Cogalois Theory in proving some very classical results in Algebraic Number Theory. More precisely, we apply our approach t o establish very easily a classical result claimed by Hecke (but not proved) in his book [71], related t o the so-called Hecke systems of ideal numbers. The last chapter of this first part of the volume contains links of Cogalois Theory with graded algebras and Hopf algebras. As is well-known, besides the classical (finite) Galois Theory there also exists its infinite counterpart, which involves the Krull topology on the Galois group of an infinite Galois extension. So, dually, it seems perfectly natural to ask about the possibility of extending the Finite Cogalois Theory to the infinite case. This is exactly the main purpose of the second part of this monograph: to develop an Infinite Cogalois Theory, that is, to investigate radical extensions E / F , not necessarily finite, with A-Cogalois correspondence. This theory includes as a very particular case the classical infinite Kummer Theory, and has nice connections with the infinite Galois cohomology and with the Pontryagin duality. This second part of the monograph is based on the author's recent papers [8], 191, [lo], [15], and [25]. In the finite case, the main tool for the investigation performed in Part I of finite field extensions with A-Cogalois correspondence was the Kneser Criterion, and the main result was a very useful characterization of these extensions in terms of a "local purity" (the n-Purity Criterion). For the infinite case we have to find the right definitions of the concepts of infinite Kneser extension, infinite Cogalois extension and infinite strongly Kneser extension, t o prove an Infinite Kneser
Criterion, and to provide a similar characterization of infinite extensions with 4-Cogalois correspondence. Infinite G-Kneser extensions are introduced and characterized in Chapter 11. One says that a field extension E / F , which is not necessarily finite, is G-Kneser if G is a subgroup of E* containing F* such that E = F ( G ) and the factor group G / F * is a torsion group having a set of representatives which is linearly independent over F . We show that the Kneser Criterion [77], which characterizes the finite separable G-Kneser extensions, can be generalized to infinite extensions. Then, we introduce and investigate infinite strongly G-Kneser extensions. The key result of this chapter is a result, which roughly speaking, states that in a tower of fields F C K 2 E , whenever two subextensions are Kneser extensions, then so is the third one. Chapter 12 is devoted to the study of infinite G-Cogalois extensions. We show that a series of results from Part I established for finite G-Cogalois extensions also hold for arbitrary infinite G-Cogalois extensions. The main tool in investigating infinite G-Cogalois extensions is the General Purity Criterion characterizing these extensions, which extends the n-Purity Criterion for finite G-Cogalois extensions. As in the case of finite extensions, we show that if an infinite extension E/F is simultaneously G-Cogalois and H-Cogalois, then necessarily G = H, and so, it rnakes sense t o define the Kneser group K n e ( E / F ) of a G-Cogalois extension E/F as the group G I F * . When the Kneser group of a G-Cogalois extension EIF has finite exponent n, we say that E / F is an n-bounded G-Cogalois extension. For such extensions, the General Purity Criterion becomes the Infinite n-Purity Criterion and has the same formulation as that of the n-Purity Criterion for finite G-Cogalois extensions. As remarked in Chapter 2, a finite extension E/F is a Cogalois extension precisely when it is T(E/F)-Kneser. This suggests that we define an extension E/F, which is not necessarily finite, t o be Cogalois if it is T(E/F)-Kneser. A series of properties of finite Cogalois extensions are shown to hold also in the infinite case. In particular, the Greither-Harrison Criterion characterizing the finite Cogalois extensions in terms of purity remains valid for infinite extensions. The prototype of infinite G-Cogalois extension is, by Kummer Theory, any classical infinite Kummer extension. We show in Chapter 13 that the essential part of the classical infinite Kummer Theory can be immediately deduced from our general approach developed in Chapter 12 by using the
10
Introductiori
Introduction
11
Infinite n-Purity Criterion proved in Section 12.1. Moreover, this criterion allows us to provide large classes of infinite G-Cogalois extensions which generalize or are closely related to classical Kummer extensions: generalized n-Kummer extensions, n-Kummer extensions with few roots of unity, and quasi-Kummer extensions. The prototype of an n-Kummer extension I i E I))/Q, where with few roots of unity is any extension of form Q ({ {a i I i E I) is an arbitrary nonempty set of strictly positive rational numbers. We derive from our Infinite Cogalois Theory results on infinite generalized n-Kummer extensions, and, in particular, on infinite n-Kummer extensions with few roots of unity, which are very similar t o the known ones for classical infinite n-Kummer extensions. Since the n-Kummer ex3, no tensions with few roots of unity, in general, are not Galois if n other approach (e.g., via Galois Theory, as in the case of classical Kummer extensions) is applicable.
infinite case, the first one is isomorphic t o the group Ch (Gal( E / F ) ) of characters of the profinite group Gal ( E I F ) . Each chapter ends with a section providing some historical notes, bibliographical comments, and credit for the material presented in that chapter. Also, each chapter includes a collection of exercises referring to, or extending the subject matter of the chapter. They contain many concrete examples and counterexamples. Many of the author's results are found as exercises. The exercises serve also as a way to draw attention t o related results in the literature.
>
Chapter 14 presents the basic facts on profinite groups, Infinite Galois Theory, character group, and Pontryagin duality which are needed in the next chapter. Chapter 15 is devoted to the investigation of infinite Galois G-Cogalois extensions. The rnost relevant results are those for n-bounded extensions. It is shown that if E/F is an infinite Galois n-bounded G-Cogalois extension then the Kneser group K n e ( E / F ) of E/F is isomorphic t o the group 2: (Gal ( E /F ) ,p, ( E ) ) of all continuous crossed homomorphisms of the compact topological group Gal ( E / F ) (endowed with the Krull topology) with coefficients in the discrete group p,,(E) of all n-th roots of unity contained in E . A similar result holds for an arbitrary G-Coga,lois extension, with p,(E) replaced by a certain subgroup p G ( E ) of the group p ( E ) of all roots of unity contained in E. This allows us to generalize t o infinite extensions a series of results from Chapter 6 concerning the investigation of Galois G-Cogalois extensions by means of crossed homomorphisms. Kext, we present the basic terminology, notation, and facts concerning lattice-isomorphic groups, which will be needed in the next section investigating infinite Abelian G-Cogalois extensions. A prototype of these extensions is any infinite classical n-Kummer extension. We are especially interested in finding the connection between the Kneser group Kne ( E I F ) and the Galois group Gal ( E / F ) of an arbitrary Abelian G-Cogalois extension E I F . In the case of finite extensions, by Theorem 5.2.2, these two groups are isomorphic, but not in a canonical way. We show that in the
Part 1
FINITE COGALOIS THEORY
CHAPTER 1
PRELIMINARIES
This chapter aims t o present some basic terminology, notation, and facts which will be freely used throughout this monograph, including a short review of Field Theory, the Vahlen-Capelli Criterion dealing with the irreducibility over an arbitrary field of binomials Xn - a, and general facts on bounded Abelian groups.
1.l. General notation and terminology
In this section we present some general notation and terminology concerning numbers, sets, groups, rings vector spaces, and modules which will be frequently used throughout this monograph.
N W
Z
= =
=
=
= =
{0,1,2,. . . ) = the set of all natural numbers N\{O) the set of all rational integers the set of all rational numbers the set of all real numbers
= the set of all complex numbers
S \ (0) for any S C_ C = {xESIX~O for) any S E R = {xES~X>O forany ) SCR m divides n
= the greatest common divisor of m and n
lcm(m, n) = the least common multiple of m and n q ( n ) = the Euler function of n
Chapter 1
1.1 General notation a n d terminology
{P E
IP
prime
1
N
the quotient group of the group G modulo H a G the quotient group Z l n Z of integers modulo n the symmetric group of degree n ord(g) = the order of an element g E G the set of all elements of G having finite order the set of all elements of G having order a power
( P \ P ) ) U (4)
{ p I p E P, p i n ) for any n E
{ m I m E N, m l n ) for any n E N
P n ID7, for any n E N

the cardinal number of an arbitrary set M the identity map on the set M the restriction of a map f : X + Y t o A
CX
the (external) direct product of an arbitrary family Cnless otherwise stated, G will denote throughout this monograph a. multiplicative group with identity element e.
1
(Gi)iI of groups
{
H C G H < G y (mod H ) H a G HVK the group with only one element H is a subset of G H i s a subgroup of G x-' E H H is a normal subgroup of G the subgroup generated by H U K the subgroup generated by
1 1Gi I xi = ei for all but finitely many i E I }

iI
the (external) direct sum of an arbitrary family
V Hi
iE 1
= the internal direct sum of an independent
family (Hi)iEI of normal subgroups of G A family (Hi)iEI of subgroups of G is said t o be independent if Hi H j ) = {e) for every i E I .
V Hi
i, I
UHi
~ 1
n (vjEI,jil
ST" Subgroups (G) (M)
{xn~xES)forany B # S C G and n E N the lattice of all subgroups of G the subgroup of G generated by the subset MCG the subgroup of G generated by the subset
For a family (Hi)ier of subgroups of G we have G = $,,,Hi only if the following two conditions are satisfied.
if and
(a) hihj = hjhi for all i, j E I , i # j and hi E Hi, h j E H j . xi, (b) Every element x E G can be written uniquely as x = where xi E Hi for all i E I, and xi = e for all but finitely i E I .
niEI
( G : H) XH
EG { g ~ , . . ,gn) . the index of the subgroup H of the group G the left coset { xh 1 h E H ) of x E G modulo H < G
Gi of a family ( G i ) i ~ of Xote that the (external) direct sum BiE1 Abelian groups is precisely the direct sum of this family in the category Ab of all Abelian groups, but not in the category Gr of all groups. Also, note that the internal direct sum $,,,Hi of an independent family (Hi)iEl of norrnal subgroups of G is canonically isomorphic t o the (external) direct sum Bi,, Hi of the family (Hi)iEl of groups. This
18
Chapter 1
1.2 A short review of basic field theory
19
allows us t o identify occasionally the internal and the external direct sum of the independent farnily (Hi)iEl of normal subgroups of a G , t o denote H i , and just to call them the direct sum of the thern identically by Hi's. The appropriate interpretation of $ can always be inferred from the context. An alternative terminology for the internal direct sum of a family of normal subgroups of a group is that of internal direct product, or of restricted direct product. Any finite Abelian group is an internal direct sum of finitely many cyclic subgroups. Since a group is cyclic if and only if it is isomorphic to Z, for some n E N, it follows that for any finite Abelian group G there exist T, n l , . . . ,n, E N* such that
eiE1
{
1< i < n
TiXi
I r l , . . . ,r,
E R ) = the left ideal of R
Moreover, one can choose the integers n l , . . . ,n, such that n l I n2 I . . . I n,. The Chinese Remainder Theorem states that if m, n E N* are relatively prime integers, then there exists a canonical group isomorphism (which is also a ring isomorphism) This implies that if n = p:l . .ap:. is the decomposition of an n E N, n 2 2 into a product of distinct prime nurnbers p l , . . . ,p,, then
a .
generated by xl , . . . ,x, E R Rx = the principal left ideal of R generated by x E R the polynomial ring in the indeterminates X I , . . . ,X, with coefficients in the ring R the degree of a polynomial f E R[X] the field of quotients of the domain R Q(FIX1, . . . ,X,]) = the field of rational fractions in the indeterminates X I , . . . ,X, with coefficients in the field F
1.1.4. VECTOR SPACES
A N D MODULES
By a set of representatives of the quotient group G / H we mean any subset S of G consisting of precisely one representative of each (left) coset modulo H .
V is a vector space over the field F FV M is a left module over the ring R RM dimF(V) = the dimension of the vector space p V
1.2. A short review of basic Field Theory
All rings considered in this monograph are unital rings, all subrings of a given ring R contain the identity element of R , and all ring morphisms preserve the identity elements. By overring of a ring R we mean any ring which includes R as a subring. A* = A \ (0) for any subset A of a ring R U(R) = the group of all units of a ring R IF = the finite field of q elements
The aim of this section is t o provide a short review of basic terminology, notation, and results in Field Theory, which will be used throughout this monograph. Recall that a field is a commutative unital ring F with 0 # 1, and such that any nonzero element of F is invertible, in other words, U ( F ) = F*. Throughout this monograph F denotes a fixed field and R a fixed algebraically closed field containing F as a subfield. Any algebraic extension of F is supposed to be a subfield of R.
20
Chapter 1
1 . 2 A short review of basic field theory
21
For an arbitrary nonempty subset S of R and a natural number n 2 1 we shall use the following notation:
and only if its Frobenius morphism p~ is actually an isomorphism.
S* = S\{O),
,!JTL(S) = { X E S ~ X ' ~ = ~ ) , sn = { x n l x E S ) . If x E R*, then 2 will denote the coset x F * of x in the quotient group 1.2.2. DEGREEO F
A FIELD EXTENSION
R*/F*.
We are now going t o review the basics of Field Theory we will currently use throughout this monograph.
Characteristic. The characteristic of a field F is a natural number defined as follows by means of the order n E W* U {oo) of the identity element I of F in the underlying additive group (F, +) of the field F :
Char ( F ) =
n 0
if EN if n=oo.
Field extension. By overfield of a field F we mean any field which includes F as a subfield. A field extension is a pair (F, E ) of fields, where F is a subfield of E (or E is an overfield of F ) , and in this case we shall write E / F . Very often, instead of "field extension7' we shall use the shorter term "extension". If E is an overfield of a field F we will also say that E is an extension of F . Of course, for a given extension E I F , F is a subgroup of the underlying Abelian group ( E , +) of the field E , so it makes sense t o consider its quotient group modulo F. According to the standard notation, this quotient group should be also denoted by E / F , which could produce confusion. To avoid that, we will make clear in the text whenever EIF has another meaning than that of a field extension, and so, no danger of confusion concerning this notation could occur. F-morphism, F-isomorphism, Galois group. If E/F and L / F are two extensions, then an F-morphism from the field E into the field L is any ring morphism u : E -+ L that fixes F pointwise, i.e., g ( a ) = a for all a E F . Any F-morphism of fields is necessarily an injective map. An F-isomorphism is a surjective (hence bijective) F-morphism. An Fautomorphism of E is any F-isomorphism from E into itself. The set of all F-automorphisms of E is a group under the binary operation of composition, called the Galois group of the extension E / F , and denoted by Gal ( E / F ) . Intermediate field, subextension, quotient extension. If EIF is an extension, then any subfield K of E , with F C K is called an intermediate field of the extension E / F . A subextension (resp. quotient extension) of the extension E/F is any extension of the form K / F (resp. E/ K ) , where K is an intermediate field of the extension E / F . If EIF is any extension and K is any intermediate field of E I F , then both K / F and E/ K are again extensions, and conversely, if K / F and E / K are extensions, then so is E / F . Lattice of subextensions. By Subextensions ( E I F ) we will denote the
If Char(F) > 0, then it is necessarily a prime number.

Characteristic exponent. The characteristic exponent e ( F ) of a field F is defined by
A field F is called perfect if F = ~ " 0 i.e., , if every element of F is an e(F)-th power in F .

Prime field. The prime subfield P ( F ) of a field F is the intersection of all subfields of F. A prime field is a field having no proper subfields, or equivalently, a field which coincides with its prime subfield. The concepts of characteristic and prime subfield are related as follows:
if P(F) G E. if P ( F ) G 0. Frobenius morphism. For any field F of characteristic p Char ( F ) =
{ ;> o
> 0, the map
is a field morphism, called the Frobenius morphism of F . As any field morphism, it is injective. So, a field F of nonzero characteristic is perfect if
set of all subextensions of E / F . Note that Subextensions ( E / F ) is a poset,
22
Chapter 1
25
that is, a partially ordered set, with respect to the partial order as follows: K is a subfield of L. KIF LIF Actually, this poset is a complete lattice, where
< defined
<
of fields, where n 2 2 and Ei is a subfield of Ei+] for every i = 0, . . . ,n - 1 is called a tower of fields.
Tower Law. Let F C K C E be a tower of fields. If B is a basis of b.K and C is a basis of K E , then { x y l x E B , y E C ) is a basis of b,E, hence
[ E : F ] = [ E : K ] . [ K : F]. In particular, the extension EIF is finite if arid only if both extensions E I K and K I F are finite. and F ( U i c l K i ) is the subfield of E obtained by adjoining the set Uio Ki t o the field F (see 1.2.3). Note that the lattice Subextensions ( E I F ) is essentially the same with the lattice Intermediate ( E / F ) of all intermediate fields of the extension EIF.
Degree. If E/F is an extension, then the underlying additive group ( E l +) of the field E may be viewed as a vector space F E via the scalar multiplication F x E + E , (a,x) I+ ax, where a x is the product of elements a E F and x E E under the given multiplication on E . In fact, this structure of F-vector space on E together with the given ring structure on E endow E in a natural way with a structure of F-algebra. We now define the degree of an extension E/F t o be the dimension of the vector space F E , and we denote it by [ E : F 1. Thus,
Ring adjunction. Let E be a ring, and let R be a subring of E . As in 1.2.2, E becomes in a natural way a commutative unital R-algebra. If A is any subset of E, then denote by R[A] the smallest subalgebra of E containing A as a subset, that is,
[ E : F ] = dim (FE).
Finite extension. One says that EIF is a finite extension (resp. an infinite extension) if [ E : F ] is a finite (resp. an infinite) cardinal. Note that for an extension E/F one has
SA = { T I A E T, T is an R-subalgebra of E ). Since we made the convention that any subring of a unital ring must contain the identity element of the given ring, it follows that a subset T of E is an R-subalgebra of E if and only if T is a subring of E which contains R as a subset. Thus,
T, T is a subring of E ) , and so, R[sl] is also the smallest subring of E containing both A and R as subsets. We call R[A] the subring of E obtained by adjoining to R the set A. or the R-subalgebra of E generated by A. The procedure to obtain R[L4] from a ring R and a subset A of an overring E of R is called ring udjunction.
Finitely generated algebra. An overring E of a ring R is said to be a finitely generated R-algebra, or an R-algebra of finite type, if E = R[A] for some finite subset A of E . If A = {a,, . . . ,a,,), then instead of R[{al,. . . , a,)] we simply write R [ a l , . . . ,a,]. Description of R[A]. Clearly, R[D] = R. For any nonempty finite subset { a l , . . . ,a,) of an overring E of a ring R, one has
where
SA = { TI A U R
An extension E/F is called quadratic (resp. cubic, quartic) if [ E : F] = 2 (resp. [ E : F ] = 3 , [ E : F ] = 4 ) . An algebraic number field, or for short a number field, is any subfield K of the field C of complex numbers, such that the extension K l Q is finite. The positive integer [ K : Q]is called the degree of the number field K . The number field K is said t o be quadratic (resp. cubic, quartic) if so is the extension KlQ.
Tower of fields. A finite chain
24
Chapter 1
25
where f ( a l , . . . ,a,,) is the "value" in ( a l , . . . ,a,) of the polynomial f . Now, if A is an arbitrary subset of El then R[Al =
U WI,
CEFA
Connection between field adjunction and ring adjunction. For any field extension EIF and any subset A of E , F(A) is the field of quotients of the domain F[A]. Commutativity of field adjunction. Let EIF be an extension, and let A, B be subsets of E . Then F ( A ) ( B ) = F ( B ) ( A ) = F ( A U B). Compositum. Let EIF be an extension, and let (Ki)iEl be a family of intermediate fields of E I F . The compositum of (Ki)iE1 is the field F ( U, , K,), denoted by ViEI Ki. For the compositum of a finite family (Ki)l< i < 7 1 we shall also use the notation K1K2 . K n .
a
where FAdenotes the set of all finite subsets of A. Commutativity of ring adjunction. Let E be a ring, let R be a subring of El and let A, B be subsets of E . Then R[A][B]= R[Bj[A] = R[A U B]. Field adjunction. Let EIF be a field extension, and let A be a subset of E . We denote by F ( A ) the smallest subfield of S containing both A and F as subsets, that is, F(A) = where A U F C K , K isasubfieldof E ) . We call F ( A ) the subfield of E obtained by adjoining t o F the set A, and the extension F ( A ) / F is called the subextension of EIF generated by A. The procedure of obtaining F ( A ) from a field F and a subset A of an overfield E of F is called field adjunction. Finitely generated extension. An extension EIF is said t o be finitely generated or of finite type if E = F ( A ) for some finite subset A of E. If A = {al, . . . ,a,,), then instead of F({al, . . . ,a,)) we simply write F ( a , . . ,an).
1 .
K,
KEEA
Linearly disjoint extensions. Let EIF be an extension, and let K l IF, K p I F be subextensions of EIF. One says that the extensions K1IF and K 2 / F are linearly disjoint, or that the fields K 1 and K 2 are linearly disjoint over F , if the canonical morphism of F-algebras is injective. In this case we have K l K 2 = K , and every linearly independent subset of K1 (resp. K2) over F is linearly independent over K2 (resp. K1). Conversely, if there exists a vector space basis of K 1 over F which is linearly independent over K 2 , then the extensions K1IF and K 2IF are linearly disjoint. The extensions K1IF and K 2 / F are linearly disjoint if and only if there exists a vector space basis Bi of K i over F , i = 1, 2, such that B 1 Bz = { xy I x E B1, y E B2 ) is a vector space basis of K1K 2 over F . Degree of compositum and linearly disjointness. Let EIF be an extension, and let K1I F , K p / F be two finite subextensions of E I F . Then, the following statements hold. [ K 1 : F] - [Kp : F ] , with equality if and only if K1 and K 2 are linearly disjoint over F . (b) If [ K 1 : F ] and [Kp : F] are relatively prirne, then K1 and K 2 are linearly disjoint over F .
(a) [ K 1 K 2 : Fj
GA = { K
Simple extension. An extension E/F is said to be simple if there exists a E El called a prinlitis~e element of EIF, such that E = F ( a ) . Description of F(A). Clearly, F ( 0 ) = F . For any nonempty finite subset {all. . . ,a,) of an overfield E of a field F one has
<
Now, if A is an arbitrary subset of El then

1.2.4. ALGEBRAIC EXTENSION
where FAdenotes the set of all finite subsets of A.
Algebraic element, transcendental element. Let E/F be an extension, and let u E E . Then u is said t o be an algebraic element over F ,
26
Chapter 1
27
or an algebraic element of the extension E/F if there exists a nonzero polynomial f E F [ X ] such that f (u) = 0. The element u is said t o be a transcendental element over F , or a transcendental element of the extension E/F if it is not algebraic over F , that is, if there exists no nonhero polynomial f E K [ X ] such that f (u) = 0. An algebraic number is any number z E C which is algebraic over Q.
Minimal polynomial. Let E/F be an extension, and let u E E . Then,
Algebraic extension. Let E/F be an extension. We say that E/F is an
algebraic extension, or that E is algebraic over F , if every element of E is algebraic over F; otherwise we say that EIF is a transcendental extension, or that E is transcendental over F. We list below some of the main properties of algebraic extensions.
1. An extension E/F is algebraic if and only if every subring R of E with F & R is a field. 2. The following assertions are equivalent for an extension E / F . (a) EIF is a finite extension. (b) E/F is algebraic and finitely generated. (c) There exist finitely many algebraic elements u l , . . . ,u,, of E/F such that E = F ( u l , . . . ,u,,). 3. Let E/F be an extension, and let A be a subset of E consisting of algebraic elements over F . Then F ( A ) / F is an algebraic extension, and F[A] = F(A). 4. Let F C K E be a tower of fields. Then E/F is an algebraic extension if and only both E I K and K I F are algebraic extensions. 5. Any F-morphism a : E -+ E of an algebraic extension EIF is necessarily an F-automorphism of E . 6. For any extension E / F , denote by F E the set of all algebraic elements of E over F . Then FE is a subfield of E containing F, which is called the algebraic closure of F in E .
the evaluation map at u defined as
e , : F [ X ] --+ E ,
4 f ) =f is a ring morphism, with image
(4
By the Fundamental Theorem of Isomorphism for rings, e, induces a ring isomorphism F[X]/Ker(&,) E F[u]. Clearly, we have u is transcendental over F u Ker(e,) = 0 w F [ X ] Z F[u], u is algebraic over F u Ker(e,) # 0. Consequently, since F [ X ] is an UFD, it follows that u is algebraic over F if and only if the ideal Ker(e,) of F[X] is generated by a unique rnonic polynomial, which is called the minimal polynomial of u over F and is denoted by Min(u, F ) . The following statements are equivalent for an algebraic element of the extension E / F and a polynomial f . (a) f = hlin(u, F ) . (b) f is a rnonic polynomial in F [ X ] of least degree such that f (u) = 0. (c) f is a rnonic irreducible polynomial in F [ X ] such that f (u) = 0. (d) f is a rnonic polynomial in F [ X ] such that f (u) = 0, and f 1 g for every polynomial g E F [ X ] with g(u) = 0.
Degree of an element. Let u be an algebraic element of an extension EIF, and let n = deg(Min(u, F ) ) . Then
1.2.5. SPLITTING FIELD

Existence and uniqueness. Let F be a field, and let f E F [ X ] \ F . A
splitting field of f over F is an overfield E of F , such that the following two conditions are satisfied.
(aj f splits over E , i.e., we can write
(a) { 1, u, . . . ,un-l ) is a basis of the vector space F[u] over F . (b) F[u] is a field, and so, F[u] = F ( u ) . (c) [ F ( u ) : F ] = n, and n is called the degree of u over F .
where X I , .. . ,x, E E and c E F * . (b) The field E is minimal with the property (a), i.e., there is no proper subfield El of E containing F such that f splits over El, or equivalently, E = F ( x l , . . . ,x,). For any field F and any polynomial f E F [ X ]\ F there exists a splitting field of f over F , which is uniquely determined up to an F-isomorphism.
28
Chapter 1
1 . 2 A short, review of basic field theory
29
Finite field. If F is a finite field, then Char(F) is necessarily a prime , .Since number p > 0, and so, its prime subfield P ( F ) is isomorphic t o Z F 1 = prL. F / P ( F ) is a finite extension, say [ F : P ( F ) ] = n , it follows that I Conversely, for every prime number p > 0 and every positive integer n , there exists a field F with (FI = pn, namely the splitting field of the polynomial XP" - X E Zp[X], and any such field F is a splitting field of the polynomial XP" - X over P ( F ) . In particular, two finite fields F and F' are isomorphic if and only if 1 F I = I F' I. Thus, for any power q of a prime number there exists a field, which we will denote by IFq, with (IF, I = q, and is unique up to an isomorphism. To uniform the notation, we will denote the field Z p of integers modulo a prime number p by IFp. For any finite field IFq and any n E W* we always will consider that IFq is a subfield of IFqn .
1.2.6. AI,GERRAICAI,T,Y CLOSED EXTENSION
(2) If E/F is an algebraic extension, then every F-automorphism of E can be extended to an F-automorphism of F . 1.2.7. ~ ' O R M A I ,EXTENSION
Conjugate elements. Let F be a field, let F be a fixed algebraic closure of F , and let x, y E F . We say that x and y are conjugate elements over F if one of the following equivalent conditions is satisfied.
(a) There exists an F-automorphism a of such that a ( x ) = y. (b) There exists an F-isomorphism T : F ( x ) + F ( y ) such that T(X)= y. (c) Min(x, F ) = Min(y, F ) .
Algebraically closed field. A field F is said to be algebraically closed if it satisfies one of the following equivalent conditions. (a) Every nonconstant polynomial in F [ X ] splits over F . (b) Every nonconstant polynomial in F[X] has a t least a root in F . (c) Every irreducible polynomial in F [ X ] has degree 1. (d) The field F has no proper overfield E such that E/F is algebraic. Steinitz's Extension Theorem. Let EIF be an algebraic extension, let R be an algebraically closed field, and let a : F -+ R be a field morphism. Then a can be extended t o a field morphism T : E --+ R. Algebraic closure. An algebraic closure of a field F is an overfield E of E such that E is algebraically closed and E/F is an algebraic extension. A classical result due to Steinite asserts that any field F has an algebraic closure, which is unique up to an F-isomorphism. An algebraic closure of F will be denoted by F . From Steinitz's Extension Theorem follow immediately the following two important facts. (1) If F is a fixed algebraic closure of F, and E/F is any algebraic extension, then there exists an F-morphism T : E + F , which is necessarily injective, and extends the canonical injection j : F r F . Identifying E with r ( E ) , we can assume that any algebraic extension of F can be considered as a subfield of F. This assumption will be made tacitly throughout the monograph.
Equivalent definitions. An extension EIF (with E contained in F, as always assumed) is said to be normal or quasi-Galois if it is algebraic, and satisfies one of the following equivalent conditions. (a) Whenever f is an irreducible polynomial in F[X], then either f splits over E, or f has no roots in E . (b) The minimal polynomial of each element of E splits over E. (c) For each x E E , all the conjugates of x over F belong to E . (d) a ( E ) & E for every F-morphism T : E -+ F . (e) aIE E Gal(E/F) for every a E Gal(F/F). Properties. We list below some of the basic properties of normal extensions.
1. Let F C K & E be a tower of fields. If EIF is a normal extension, then so is also E I K . 2. Let (Ej/F)jEI be a nonempty family of normal extensions. Then E,) IF and F(U~,, Ej)IF are also normal extensions, i.e., the meet and the compositum of any family of normal extensions is again normal. 3. For any algebraic extension E/F there exists a "least" normal extension E/F containing E/F as a subextension (and, of course, in F ) , where E is the intersection of all subfields N of contained F containing E such that N / F is a normal extension. The extension E/F is called the normal closure of the extension E I F . It can be described as follows: let A be a subset of E which generates the extension E I F , i.e., E = F ( A ) , e.g., A could be E itself. Denote by B the set of all conjugates (in F ) over F of all elements in A. Then
(n,,
Chapter 1
31
= F ( B ) . This shows that the normal closure of a finite extension is also a finite extension.
A polynomial f E F [ X ] is said to be separable if all its irreducible factors in F [ X ] are separable.

Separable element, separable extension. Let E/F be an extension.
1.2.8. SEPARABLE EXTENSION

Multiple root. Let F be a field, and let f E F [ X ] \ F have a root a E F .
Then
f = (X-a)".g for some m E W* and some g E F [ X ] with g(a) # 0, which are both uniquely determined. The number m is called the multiplicity of the root a. We say that a is a multiple root if m > 1 and a simple root if m = 1.
Derivative. For any polynomial f = a0
+ a l X + . . . + a n X n E F [ X ] with
An algebraic element u of E/F is said to be separable over F if its minimal polynomial Min(u, F ) is separable over F . An algebraic extension E/F is said to be separable if every element of E is separable over F . Though the concept of separability is more general and applies for instance to arbitrary algebras over commutative rings, in particular t o arbitrary field extensions, we shall deal in this monograph only with algebraic separable extensions. So, a separable extension EIF means for us an algebraic separable extension.
Separable degree. Let F be a field of characteristic p > 0, and let f ( X ) E F [ X ] be an irreducible polynomial. Then, there exists a unique number e E W such that f ( X ) E F[XP'] but f ( X ) # F[x~'+'], and write f ( X ) = ~ ( x P " )for some g ( X ) E F [ X ] . The following statements hold.
coefficients in a field F we define the derivative D f of f by the rule where, as usually, kak = ak called differentiation
+ . . . + ak
(k times). Thus, one obtains a map,

I+
D : F [ X ] +F [ X ] , f which has the following properties.
Df,
(1) D ( f + g) = D f + Dg, (2) D b f ) = a D f , (3) D ( f g) = ( D f )g + f (Dg)3 for every f , g E F [ X ] and every a E F . Differentiation provides a useful test for multiple roots: an element a E F is a multiple root of a polynomial f E F [ X ] if and only if f (a) = ( D f (a) = 0. Let F be a field of characteristic p 2 0, and let f E F[X]. Then D f = 0 if and only if f E F[XP]. If p = 0, then f E F[Xp] means exactly that f E F , and if p > 0, then f E F[XP] means exactly that there exists g ( X ) E F [ X ] such that f ( X ) = g(Xp).
Separable polynomial. Let F be a field. An irreducible polynomial f E F [ X ] is said to be separable over F if one of the following equivalent conditions is satisfied. (a) In every splitting field of f over F , f has only simple roots. (b) In some splitting field of f over F , f has only simple roots. ( 4 D f # 0. (d) Either Char(F) = 0, or Char(F) = p > 0 and f # F[XP].
(a) g ( X ) is an irreducible and separable polynomial over F . (b) All roots of f ( X ) in F have the same multiplicity equal t o p", and deg(g(X)) is equal t o the number of distinct roots in F of f ; in particular
d e d f (XI) = pe . deg(g(X)). The positive integers deg(g) and pe are called the separable degree of f and the degree of inseparability of f , respectively. If E/F is an extension and u E E is an algebraic element, then one defines the separable degree of u over F (resp. the degree of inseparability of u over F ) as being the separable degree of Min(u, F ) . If E/F is a finite extension, then the separable degree of E/F is defined t o be the (finite) number, denoted by [ E : F], , of all F-morphisms from E into F. The connection between the numbers considered above is the following. Let F be a field of characteristic p > 0, let E/F be an extension, and let u E E be an algebraic element over F having the degree of inseparability p". Then [ F ( u ) : F], is equal to the separable degree of u over F, and [ F ( u ) : F] = pe - [ F ( u ) : F ] , .
Properties. We list below some of the basic properties of separable ex-
tensions.
32
Chapter 1
I. A field F is perfect if and only if every algebraic extension of F is separable. In particular, any algebraic extension of a field of characteristic 0 or of a finite field is a separable extension. 2. Let E/F be any extension, and let A be a subset of E consisting of algebraic separable elements over F . Then F ( A ) / F is a separable extension. 3. Let F E K E be a tower of fields. Then E/F is a separable extension if and only if both K / F and E I K are so. 4. Let EIF be any algebraic extension and denote by E, the set of all elements of E which are separable over F . Then E,/F is a subextension of EIF, which is the greatest separable subextension of E I F . The subfield E, of E is called the separable closure of F in
we define the following elements:
E. 5. For any finite extension E I F , one has [ E : F ] , = [E, : F]; in particular, EIF is separable if and only if [ E : F], = [ E : F]. .Also [ E : F 1, I [ E : F 1, and the positive integer
[ E : FIi = [ E : F ] / [ E : F j , is called the inseparable degree of EIF. Observe that [ E : FIi = 1 if Char(F) = 0, and [ E : FIi = pm for some m E N if Char(F) = p. 6. Let ( E j / F ) j E I be a nonempty family of separable extensions. Then F( U j E I Ej)/F is also a separable extensions, i.e., the compositum of any family of separable extensions is again separable. 7. Any finite separable extension is simple.
Purely inseparable element. Let E/F be an extension, where F is a field of characteristic p > 0. An element u E E is said t o be purely inseparable over F if upm' E F for some m E N; if e E N is the least such m , then Min(u, F) = XP' - upe. This immediately follows from the following classical result due t o Abel: if a E F \ F p , then the polynomial XP" - a is irreducible over F for every n E N. An element u E E is simultaneously separable and purely inseparable over F if and only if u E F . Purely inseparable extension. An extension E/F is said t o be purely inseparable if every element of E is purely inseparable over F . For any algebraic extension E / F , the extension E,/F is separable and the extension E/Es is purely inseparable. Norm and trace. Let E/F be a finite separable extension of degree n , and let 01, . . . ,a, be all distinct F-morphisrns of E into F. For any a E E
which are called the n o r m and the trace of a in the extension E I F , respectively. A priori, the elements NEIF(a) and TrsiF(a) are in F , but actually, they both belong to F . The map NEIF : E + F , called the n o r m from E t o F, or the n o r m of EIF, is multiplicative, and the map TrEII' : E + F, called the trace from E t o F, or the trace of E / F , is F-linear. More generally, the concepts of "Norm" and "Trace" can be defined for any pair of commutative rings (R, S ) , with S an overring of R such that S is a free R-module of finite rank. 1.2.9. GALOIS EXTENSION
Equivalent definitions. An extension EIF is said t o be a Galois extension if it is algebraic, and if one of the following equivalent conditions is satisfied.
(a) E/F is a normal and separable extension. (b) Every element u of E such that a ( u ) = u for all a E Gal(E/F), necessarily belongs t o F. (c) For every u E E, Min(u, F ) splits over E and has only simple roots.
Properties. We list below some of the basic properties of Galois extensions.
1. Let F E K E be a tower of fields. If E/F is a Galois extension, then so also is E / K . Gal(E/K) Gal(E/F). If E/F is a Galois extension, then K / F is also a Galois extension if and only if Gal(E/K) a Gal(E/F), and in this case the map Gal(E/F) -+ Gal(K/F), a c-, alK induces an isomorphism of groups
<
34
Chapter 1
35
2. Let (Ej/F)jE1 be a nonempty family of Galois extensions. Then Ej)/F and F(Uj,, Ej)/F are also Galois extensions, i.e., the meet and the cornpositurn of any family of Galois extensions is again Galois. 3. For any field F and any subset A of F consisting only of separable elements over F, let B be the set of all conjugates (in p) over F of all elements in A. Then F ( B ) / F is a Galois extension, which is called the Galois extension of F generated by the subset A of F . In particular, the normal closure E/F of any separable extension E / F is a Galois extension. 4. A finite extension E/F is a Galois extension if and only if E is the splitting field over F of a separable polynomial in F[X].
(njE,
(resp. decreasing, or order-reversing), i.e., satisfies the following condition: (resp. V x y in P f (y) f (x)). A poset isomorphism (resp. a poset anti-isomorphism) from a poset P into a poset PI is a bijective poset morphism (resp. poset anti-morphism) f : P --+ P I such that its inverse f-' : P' + P is a poset morphism (resp. poset anti-morphism) . If L and L' are two lattices, then a lattice morphism (resp. a lattice anti-morphism) from into L into L' is a map f : L + L' satisfying the following conditions:
<
<
Fixed field. Let E be any field. For any nonempty subset S of the group of automorphisms of the field E we denote by Fix(S) or by E~ the fixed field of S , that is, Fix(S) = { x E E I u(x) = x , V a E S ) , which is a subfield of E . It is also called the field of invariants of S , and denoted by Inv(S). If EIF is an extension with Galois group G and S is any nonempty subset of G, then Fix(S) contains F as a subfield. With the notation above, an algebraic extension EIF is Galois if and only if Fix(Gal(E/F)) = F. The Artin Theorem. Let E be any field and let G be any finite group of automorphisms of E . Then E/Fix(G) is a finite Galois extension, G = Gal(E/Fix(G)), and [E : Fix(G) ] = (GI. Poset, lattice, isomorphism, anti-isomorphism. A partially ordered set, or poset, is a pair (P, <) consisting of a nonempty set P and a binary on P which is reflexive, anti-symmetric, and transitive. Very relation often, a poset (P, 6 ) will be denoted shortly by P . The opposite poset of P will be denoted by Pop. A lattice is a poset L in which every two elements x, y have a least upper bound x V y (also denoted by sup(x, y) ) and a greatest lower bound x A y (also denoted by inf (x, y) ). A poset in which every subset A has a least upper bound VzEAx (also denoted by sup(A) ) and a greatest lower bound AzEA x (also denoted by inf(A) ) is called a complete lattice. -4poset morphism (resp. a poset anti-morphism) from a poset P into a poset P' is a map f : P + P' which is increasing, or order-preserving
Any lattice isomorphism between complete lattices commutes with arbitrary meets and joins. A bijection f : L + L' between two lattices L and L' is a lattice isomorphism (resp. lattice anti-isomorphism) if and only if f and its inverse f are both order-preserving (resp. order-reversing) maps, i.e., if and only if f is a poset isomorphism (resp. poset anti-isomorphism).
-'
The Fundamental Theorem of Finite Galois Theory. The following statements hold for a finite Galois extension E/F with Galois group G. (a) The maps
a : Intermediate ( E I F ) --+ Subgroups (G), a ( K ) = Gal(E/K),

and
p : Subgroups (G) -4 Intermediate ( E I F ) , B(H) = Fix(H),

establish anti-isomorphisms of lattices, inverse to one another, between the lattice Intermediate ( E / F ) of all intermediate fields of the extension E/F and the lattice Subgroups (G) of all subgroups of G. (b) For any H I H2 G one has
<
<
<
In particular, I GI = [ E : F], and [HI = [ E : Fix(H) ] and (G : H ) = [Fix(H) : F ] for any H
<G
Chapter 1
37
(c) If K E Intermediate(E/F), then K / F is a Galois extension if and only if Gal(E/K) <I G a l ( E / F ) , and in this case
Lattice-theoretical aspects of Galois Theory. Let E/F be a finite Galois extension with Galois group G, let (Hi)iEI be a nonempty family of subgroups of G, and let (Kj)jE be a nonempty family of intermediate fields of the extension E / F . Then ~ . ~ (a) ~ i x ( n Hi) = Vier Fix(Hi). I Hi) = I Fix(Hi). (b) ~ i x ( V , , Gal(E/Kj). (c) Gal(E/(VjEJ K j ) ) = ( 4 W E / ( n j E J K j ) ) = VjEJ Gal(E/K&
of intermediate fields Ei of E I F , i = 1 , . . . , n , where Ei is the fixed field of the subgroup GI c~ . . . {e) @ I . . . @ G,, of G, where the identity element e of G occurs in the i-th position. Moreover, every extension E i / F , 1 i n is Galois, and
< <
for every i, 1
< i < n - 1.
niE
nieJ
Base field change. Let EIF and K I F be subextensions of an extension L/ F . If E / ( E n K) is a finite Galois extension, then E K / K is also a finite Galois extension, and the map
Gal(EK/K)
Root of unity. Let F be a field, and let n E N*. An element C E F is said t o be an n-th root of unity (or of 1) if Cn = 1. A root of unity is an n-th root of unity for some n E N*. We denote by p n ( F ) the set of all n-th roots of unity in F, and by p ( F ) the set of all roots of unity in F . Then
for all n E W . We also have
-+ Gal(E/(E n K ) ) , a I-, alE,
is an isomorphism of groups.
Compositum of Galois extensions. Let El /F and E2/Fbe finite Galois extensions, and assume that E l , E2 are subfields of some other field. Then El E2/Fis a finite Galois extension, and the map
is an monomorphism of groups, which is an isomorphism when El nE2 = F. More generally, let ( E i / F ) l < i < n be a finite family of finite Galois extensions. Assume that all E l , . . . ,En are subfields of some other field, and Ei+,r!(El -Ei) = F for every i , 1 i n - 1. Then
< <
Conversely, whenever the Galois group G of a finite Galois extension
E/F can be decomposed into a finite internal finite direct sum of subgroups
then the field E can be expressed a s the compositum
Let n E N* . The group p,,(F) is a cyclic group having order a divisor of n. In case the polynomial X n - 1 splits over F (in particular, this happens whenever F is algebraically closed), then p,(F) is a cyclic group of order n if and only if gcd(n,e(F)) = 1. Recall that e ( F ) denotes the characteristic exponent of F (see 1.2.1). By a primitive n-th root of unity over F we mean throughout this monograph any generator of the cyclic group p n ( F ) , and Cn will always denote such an element. If gcd(n, e ( F ) ) = 1, then there exist precisely cp(n) primitive n-th roots of unity over F , where cp(n) is the Euler function of n. If there is no danger of ambiguity about the ground field F , we will simply say primitive n-th root of unity. Note that the standard definition of the notion of primitive root of unity is somewhat different: usually, by a primitive n-th root of unity over F one understands any element in F having order n in the group -* F . Clearly, when gcd(n,e(F)) = 1, then the two definitions coincide, but if Char(F) = p > 0 and pl n , then no primitive n-root of unity in the standard definition exists, while, in our definition, always exists a primitive 11-th root of unity.
38
Chapter 1
1.3 The Vahlen-Capelli Criterion
39
Cyclotomic extension, cyclotomic polynomial. Let F be a field, let n E W be such that gcd(n, e ( F ) ) = 1, and let Crl be a primitive n-th root of unity over F . The field F(Cn) is called the n-th cyclotomic field over F , and the extension F(Cn)/F is called the n-th cyclotomic extension over F . Since F(C,,) is the splitting field over F of the separable polynomial X n - 1, it follows that any cyclotomic extension is a Galois extension. . .. ,E ~ r, = ( ~ ( n )be , all primitive n-th roots of unity over F . Let ~ 1 , Then r
i= 1
Let n E N*, and let EIF be a finite extension such that gcd(n, e ( F ) ) = 1 and Cn E F . Then EIF is a cyclic extension with [ E : F ] I n if and only if there exists x E E* such that xrLE F and E = F ( x ) .
1.3. The Vahlen-Capelli Criterion The aim of this section is t o prove a criterion for the irreducibility of binomials X n - a over an arbitrary field, known as the Vahlen-Capelli Criterion. A special case of this criterion, which is mainly applicable t o subfields of R, is also presented. Throughout this section F will denote an arbitrary field and R a fixed algebraically closed field containing F as a subfield. We start with some preparatory results which are needed in the proof of the Vahlen-Capelli Criterion.
is a monic polynomial in F [ X ] , and is called the n-th cyclotomic polynomial over F.

Galois group of a cyclotomic extension. Let F be a field, and let n E N* be such that gcd(n,e(F)) = 1. Then Gal(F(&)/F) is isomorphic t o a subgroup of the group of units U(Zn) of the ring Zn. Moreover, the following conditions are equivalent. (a) Gal(F([,)/F) 2 U(Zn). (b) an is irreducible over F . (c) iF(Cn) : FI = ( ~ ( n ) Cyclotomic extension of Q. Let n E W ,let Cn be a primitive n-root of unity over Q, e.g., Cr, = c o s ( 2 ~ l n ) i s i n ( 2 ~ / n )and , let <Pn be the n-th cyclotomic polynomial over Q. Then
LEMMA1.3.1. Let a E F , and let m, n E W be relatively prime integers. Then, the polynomial Xm" - a is irreducible in F [ X ] if and only if both polynomials X" - a and X" - a are irreducible in F [ X ] . PROOF. Since Xrnrl- a = (Xn)*' - a = (X")rL - a, we deduce that if X 7lL7" a is irreducible in F[X], then so are X m - a and X n - a.
Conversely, assume that both X" - a and X n - a are irreducible in F [ X ] . Let u E R be a root of Xrnn-a. Then urn is a root of the irreducible -a, polynomial X n - a , and un is a root of the irreducible polynomial X7IZ hence [ F ( u m ): F] = n and [ F ( u n ) : F] = m. Using the Tower Law for the following two towers of fields F ( u ) and F E F ( u n )
1.2.11. ABELIAN EXTENSION,
CYCLIC EXTENSION
F(um)
F(u),
Abelian extension. An extension E I F , which is not necessarily finite, is said t o be Abelian if E/ F is a Galois extension and Gal(E1F ) is an Abelian group. Any cyclotomic extension is Abelian. Also, any subextension and any quotient extension of a cyclotomic extension is Abelian. Cyclic extension. An extension E I F , which is not necessarily finite is said t o be cyclic if EIF is a Galois extension and Gal(E/F) is a cyclic group.
we deduce that n l [ F ( u ) : F ] and m l [ F ( u ) : F ] , hence mnI [ F ( u ) : F] since gcd(m, n) = 1. But u is a root of the polynomial Xrnr' - a , SO [ F ( u ) : F] mn. It follows that [ F ( u ) : F] = mn, which shows that Min(u, F ) = Xmn-a. This proves that Xrnn-a is irreducible in F[X]. 0
<
LEMMA 1.3.2. Let p E l P and a E F . Then XP - a is irreducible in F [ X ] if and only if a @ F P .
40
Chapter 1
1.3 T h e Vahlen-Capelli Criterion
41
PROOF. If a E FP, then a = b p for some b E F, and then XP - a = XP - b p is clearly reducible in F[X]. Conversely, assume that a # FP, and prove that XP - a is irreducible in F [ X ] . Suppose the contrary, and let f be an irreducible factor of degree n, 1 Q n < p, of XP - a. If u is a fixed root of f in R, then the roots of f in R all have the form [u, where C E R is a p t h root of unity. Let c be the constant term of f . Since k c is the product of the roots of f , we have fc = <u7', where J E fl is a p t h root of unity. Since n and p are relatively prime integers, there exist r, s E Z such that rn sp = 1, and so,
NOWobserve that oi(u) # oj(u) for every i # j in (0,. . . , p - 1) since has order p in R*. This implies that ai(v) # aj(v) for every i # j in {0, . . . , p - I ) , so the polynomial f of degree p has a t least p distinct roots vo, . . . ,up-1 in E . Consequently, these elements are exactly all the roots vk E F. We now multiply together the in R of f , and then w = equalities [ku = vi, k = 0 , . . . , p - 1, and obtain qup = qa = wP, where
Cp
nizi
If p is odd, then q = 1, and so, a = wP E F P , which contradicts Lemma 1.3.2. This proves (1). Thus uJT E F , hence a = up = (uJr)p E assumption.
Fp!
which contradicts our
LEMMA 1.3.3. Let p E B and a E F be such that XP - a is irreducib~ in F [ X ] , and let u E R be a root of XP - a . (1) If p > 2, or if p = 2 and Char(F) = 2, then u @ F(u)p. if and only if -4a E F 4 . (2) If p = 2 and Char(F) # 2, then u E F ( u ) ~
PROOF. (1) Assume that u = vp for some v E F ( u ) . Then we have . . ,ck-l E F . v = Ck=O ckuk for some CO,. If Char(F) = p, then we have
. u = v2 with v = a + P u , a , P E F . (2) Assume that u E F ( u ) ~ Then Thus u = (a P u ) ~ = a2 P2a 2apu. Since (1, u) is a basis of the vector space F ( u ) over F, we deduce that
a 2 + P 2 a = 0 and P a p = 1. Eliminating P, we obtain a = -4a4, and so Conversely, if -4a E F 4 , then we can write -4a = -y4 = (2a)' for some y,a E F * . Kow, if we take p = (2a)-', then it is easily verified that u = (a + P u ) ~ EF(u)~. 0
hence a = UP E FP, which contradicts Lemma 1.3.2. In particular, if Char(F) = p = 2, then we have u $ ! F(u)p. If Char(F) # p, then consider the field E = F ( u , Cp). Since E is clearly the splitting field of the polynomial XP - a E F [ X ] , EIF is a normal i p - 1. extension. Any o E Gal(E/F) sends u into some ~ j u ,0 Conversely, for every i E {0, . . . , p - I ) , C ~ U is a root of the irreducible polynomial XP - a , hence it is a conjugate of u over F . Therefore, there exists oi E G a l ( E / F ) such that oi(u) = [ju. Set vi := oi(v), 1 i Q p-1. Then oi(vP) = = oi(u) = ~ i u . By Lemma 1.3.2, v E F ( u ) \ F, hence F ( v ) = F ( u ) . If we set f := Min(v, F ) , then
LEMMA 1.3.4. Let p E I?, n E N, n 2 2, and a E F . (1) If p > 2, or if p = 2 and Char(F) = 2, then XP" - a is irreducible in F [ X ] if and only if a # FP. (2) I f p = 2 and Char(F) # 2, then x~~ - a is irreducible in F [ X ] if and only if a # F2 and -4a # F 4 .
PROOF. (1) If a E FP, then a = bP for some b E F, hence
< < <
is divisible by
xpn-'- b.
deg(f) = [ F ( v ) : F] = [ F ( u ) : F ] = deg(Min(u, F ) ) = deg(XP - a ) = p.
Conversely, assume that a # F p . Let v E R be a root of XP" - a , n-1 and set u := UP . Then up = a , hence [ F ( u ) : F] = p by Lemma 1.3.2. Clearly XP' - a is irreducible if and only if v has degree pn over F , which will be proved by induction on n. This is true if n = 1 by Lemma 1.3.2. By Lemma 1.3.3 (I), u # F(u)P, hence the polynomial XP"-' -u E F(u)[X] is
42
Chapter 1
1.3 The Vahlen-Capelli Criterion
43
irreducible over F ( u ) by the inductive hypothesis. Consequently, the root v of xpn-l- u has degree pn-l over F(u). Then, by the Tower Law, it follows that v has degree pn over F, and we are done. (2) If a E F ~ then , clearly x2" - a is reducible over F . Now assume that -4a E F" Then -4a = c4 for some c E F . Since Char(F) # 2, we have c = 2b for some b E F, hence then If we set Y = x2"-',
THEOREM 1.3.5 (THEVAHLEN-CAPELLI CRITERION). Let F be an arbitrary field, let n E N*, and let a E F . Then, the following assertions are equivalent. (1) The polynomial X n - a is irreducible in F[X]. (2) a $ F P for all p E PT1,and a # -4F4 whenever 4 1 n.
PROOF. Of course, the result is vacuously true for n = 1, so we can assume that n 3 2. Let p be a prime divisor of n. If a E FP, then a = bp for some b E F , and XrL-a is divisible in F [ X ] by X m-b, where m = n l p , hence X7' - a is reducible in F [ X ] . - a is a reducible If 4 1 n and a = -4b4 for some b E F , then XTL polynomial in F [ X ] since
= (y2
+ 2b2)2- ( 2 b ~ = )~ ( y 2 + 2bY + 2b2)(y2- 2bY + 2b2),
hence X2n - a is reducible in F[X]. Conversely, assume that Char(F) # 2, a # F~ and -4a $ F ~ .We - a is irreducible over F by induction on n. Let v E R be prove that x2" 2" -1 . Then u2 = a , hence [ F ( u ) : F] = 2 a root of x'" - a , and set u := v -a by Lemma 1.3.2. Therefore, as in the proof of ( I ) , the fact that x2" is irreducible over F means precisely that x2'-' - u is irreducible over F(4. If n = 2, then v' = u # F ( u ) ~ by Lemma 1.3.3 (2), hence X 2 - u is irreducible over F ( u ) . If n > 2, then by the inductive hypothesis, x2"-' - u E F(u)[X] is irreducible over F ( u ) if and only if u # F ( u ) ~ and -4u 6 F ( U ) ~ Kow . observe that -4u E F ( u ) ~ =j -21 E F(u)' , since 0 # 4 = 22 E F ( u ) ~and
This proves the implication (1)
(2).
Conversely, assume that condition (2) is satisfied. Let
be the decomposition of n as a product of mutually distinct prime numbers p l , . . . ,p,, with r, kl,. . . ,k, E W .By Lemma 1.3.1, we only need to check that XP;' - a is irreducible in F [ X ] for every i, 1 $ i $ r. This follows from Lemma 1.3.2, Lemma 1.3.4, and the following simple fact: if 0 Char(F) # 2 and a # -4F4, then -4a # F 4 . The next result is an alternative version of the Vahlen-Capelli Criterion.
since the map defined by u Consequently, we have u
I +
-u is an element of Gal(F(u)/F).
THEOREM 1.3.6 ( A VARIANT OF THE VAHLEN-CAPEI,LI CRITERION). Let F be an arbitrary field, let n E N, n 3 2, and let a E F*. Then, the polynomial X " - a is reducible in F [ X ] if and only if one of the following two conditions is satisfied. (1) a E F s for some s E N , s 3 2, s [ n , or (2) 4 ( n and -4a E F * l .
PROOF. First, observe that for any a E F * , one has then , -4a = b4 for some b E F* a.nd 4 # 0, Indeed, if -4a E F * ~ hence Char(F) # 2. Thus b = 2c for some c E F * , and so, -4a = 16c4. Multiplying the last equality by (-4)-' we find that a = -4c4 E -4F*4. The implications above can be clearly reversed.
But -4a
# F ( u ) ~and - 421 # ~ ( u ) ~ u # F ( u ) ~ . # F 4 , hence u # F ( u ) ~ by Lemma 1.3.3 (2). This proves that

0
x"-'
- u is irreducible over F (u), and we are done.
We are now in a position t o state and prove the main result of this section. We shall use the nota.tion -4F4 := { -4b4 I b E F ).
46
Chapter 1
1 . 4 Bounded Abeliari groups
X k - c, we deduce that c = ei for some e E F* and i E N, i Then k = i j > j for some j E N*, and
>
I , i I k.
1.4. Bounded Abelian groups
In this section we present some basic properties of Abelian groups of bounded order which will be frequently used in the sequel. Throughout this section G will denote an arbitrary multiplicative group with identity element e. If n E W , we set Gn = { x n 1 x E G ) . For any torsion group G we will use the following notation:
p , ( F ( f i ) ) )c F . Thus pi(R) n F ( G ) = p i ( F ( f i ) ) fi' E F* with 1 j < k, which contradicts the definition of k. Consequently, Min( fi, F ) = X k - c = X m - b, hence m = k, m I n , 0 and m = ord( The proof is now complete.
~j
so
G).
<
OG = {ord(x) l x
E G).
R.EMARK 1.3.11. Assertion (1) in Proposition 1.3.10 can be restated as follows:

(1') The polynomial X n - a is irreducible in F [ X ] if and only if a
#Fp
0
for every p E IF',.
PROPOSITION 1.3.12. Any field F satisfies the condition Co(n;a ) for any n E P and any a E F * .
PROOF. If n = e ( F ) , then pn(R) = (1) and we have nothing to prove. We can therefore assume that n # e ( F ) . Since p n ( F ) = pn(R) n F* is a subgroup of the group p,(R) of prime order n, we have either p n ( F ) = p7L(fl) or p n ( F ) = (1). If p n ( F ) = pn(R) then clearly pn(R) 5 F . Now assume that p,,(F) = (1). Let @ E R be an arbitrary root in R of X'" a if a # F'" a,nd let fi = b if a = b7"or some b E F. We claim that pTL (F( fi )) # p7,( a ) . = F(b) = F , hence This is clear when a E Fn because F(*) jL7L( ( fi)) = ~ 7 l F(, = {I) # j17L (O)' Now consider the case when a # F", and suppose that p n ( F ( = prL ( 0 ) . Then Cn E F ( @ ), hence F(<,,) C F ( fi), and so
Recall that a group G is said t o be a torsion group if every element of G has finite order. If a , b E N, then gcd(a, b) (resp. lcm(a, b)) will denote the greatest common divisor (resp. the least common multiple) of a and b. For a nonempty finite set A of natural numbers, lcm(A) will denote the least common multiple of all numbers of A, and max(A) will denote the greatest number of A. DEFINITION 1.4.1. A group G is said to be a group of bounded order if G is a torsion group and the subset OG of N is a bounded set, or equivalently, a finite set. 0 Clearly, any finite group is a group of bounded order, and any direct sum or direct product of infinitely many copies of a finite Abelian group of order n > 1 is an infinite group of bounded order. The structure of Abelian groups of bounded order is given by the following classical result.
THEOREM 1.4.2. Any Abelian group of bounded order is isomorphic to a direct sum of finite cyclic groups.
PROOF. See Kaplansky [73, Theorem 61.
G))
On the other hand, [ F ( @ ) : F] = n by Lemma 1.3.2, and [F(CTL) : F] 6 p ( n ) = n - 1, where cp is the Euler function. Consequently, [ F(C,,) : F] = 1, i.e., <, E F . It follows that Cn E p n ( F ) = {I), which is a contradiction. )) # p n ( 0 ), and then, necessarily This proves our claim: prl( F( 0 p n ( F( fi)) = (1). Thus F satisfies the condition Co(n;a ) .
As is well-known, any torsion Abelian group G is the internal direct sum of its p-primary components t,(G), p E P. Using this fact, Theorem 1.4.2 implies that an Abelian group G is a group of bounded order if and only if there exists a finite subset { p l , .. . ,p,) of the set P of all positive prime numbers and a finite family (Gi)16i6T of Abelian groups, such that
r
G ~ $ G ~ and Gi E
i= 1
j E Ai
$ zPi
nij,
:,
:
F
#
where A1,. . . ,, 4 , are arbitrary sets, nij E W ,and for each i = 1 , . . . ,r , the set { n i j I j E Ai ) is a bounded subset of N*.
48
Chapter 1
1 . 4 liourlded Abelian groups
49
Using this description, the results of this section about orders and exponents become almost obvious. In fact, they are the same as if the considered Abelian groups of bounded order were finite. However, we prefer a more elementary approach which does not refer t o Theorem 1.4.2. LEMMA 1.4.3. The following statements hold for an arbitrary Abelian group G. (1) For any a E OG and any d E N with d l a one has d E OG. (2) For any a , b E OG one has lcm(m,n) E OG, and so, lcm(A) E OG for any finite nonempty subset A of OG. (3) If G is a group of bounded order and m = max(OG), then every element of OG divides m . PROOF. (1) If a E OG and d 1 a , d E N, then a = ord(x) for some x E G , hence d = ord(xul", and so d E OG. (2) First, we observe that if a , b E OG, with gcd(a, b) = 1, then ab E OG. Indeed, if a = ord(x), b = ord(y), with x, y E G , it is known that ab = ord(xy), so ab E OG. By induction one proves that if a1 , . . . ,a, E OG, with gcd(ai,a j ) = I for all i # j in (1,. . . ,r ) , then a1 - . . . . a, E OG. We are now going t o prove that if a , b E OG, then lcm(a, b) E OG. Clearly, we can suppose that a > 1 and b > 1. If f f l . . . . . p F a and b = p f l - . . : d 8 , a=p1 with pl, . . . ,p, mutually distinct primes and a l , . . . , a,, ,&, . . . ,P, E N, divides a or b, so p ~ a X ( a i l B i ) E, OG, i = 1 , . . . , S , by (1). then m a x (a,,P,) 1=4ff1141) . . . ps Consequently, lcm(a, b) = p, E OG.
and is denoted by exp(G). The group G is said to be n-bounded if G is a group of bounded order and exp(G) = n .
LEMMA 1.4.6. Let G be an n-bounded Abelian group. If G k = {e) for some k 2 1, then n 1 k . PROOF. By the Division Theorem we can write k = nq r , with q, r E N, 0 r < n. Since G k = {e) it follows that GnQ+' = {e), hence GT = {e). But n E W is minimal with Gn = { e ) , so necessarily r = 0. Thus k = nq, i.e., n I k. 0
<
PROPOSITION 1.4.7. Let G be an n-bounded Abelian group, and let m = max(OG), c = lcm(OG). Then n = m = c.
PROOF. \ve have c 1 m by Corollary 1.4.4. But m E OG, so m 1 c by the definition of c = lcm(OG). We deduce that m = c. By Corollary 1.4.4 we have G"" {e), hence n 1 m by Lemma 1.4.6. Since m E OG there exists xo E G with m = ord(xo). But xz = e and consequently ord(xo) ( n , i.e., m 1 n. From n 1 m and m 1 n it follows that El m = n. Recall that for a given n E W we denote by IDn the set of all divisors d E N of n , and by Pn the set of all divisors p > 0 of n , with p odd prime or 4.
py"(ailDi)
PROPOSITION 1.4.8. Let G be an n-bounded Abelian group. Then, for any d E N, d [ n , there exists xd E G such that d = ord(xd). In particular, for every p E P, there exists x, E G such that p = ord(x,).
PROOF. By Proposition 1.4.7 we have n = m = max(OG), and by Lemma 1.4.3 (1) we have d E OG for every a E OG and d 1 a. Since m E OG, we deduce that d E OG for every divisor d of m = n. Thus d = ord(xd) for some xd E G. 0 REMARK1.4.9. Proposition 1.4.8 can be reformulated as follows: If G is any n-bounded Abelian group, then OG = ID,. E l The connection between the exponent and the order of a finite Abelian group is given by the next result.
If A is an arbitrary nonempty finite subset of OG, then the fact that lcm(A) E OG follows by induction on IAl.
(3) Let a E OG. By (2), we have lcm(a, m) E OG, so

But m
< lcm(a, m), hence m = lcm(a, m). Thus a ] m , as desired.
COROLLARY 1.4.4. If G is an Abelian group of bounded order, and m = max(OG), then G" = {e) and lcm(OG) I m. PROOF. Apply Lemma 1.4.3.
DEFINITION 1.4.5. Let G be a group of bounded order. The least number n E N* with the property that Gn = {e) is called the exponent of G
PROPOSITION 1.4.10. For any finite Abelian group G, exp(G) divides lGl, and [GI divides a power of exp(G). In particular, [GI and exp(G) have the same prime divisors.
50
Chapter I
1.6 Exercises t o Chapter 1
51
PROOF. Set n = IGI and m = exp(G). Since clearly G7' = {e), one has m 1 n by Lemma 1.4.6. Now we shall prove that n divides a power of m by induction on n = [GI. The statement is clear for n = 1 and n = 2. Let G be an Abelian group of order n ) 3, and assume that the statement is true for any Abelian group of order < n. Pick an element g E G, g # e and set H = (g). Then [HI = ord(g) divides m = exp(G) since gm = e. Since G is an Abelian group, we can consider the quotient Abelian group G / H which has order < n. By the induction hypothesis, IG/HI divides a power of exp(G/H). On the other hand, exp(G/H) divides exp(G) by Lemma 1.4.6, [HI divides exp(G), and n = [GI = IG/ H I . I H 1, hence n divides a power of m , as desired. 0 COROLLARY 1.4.11 (CAUCHY'S LEMMA). For any finite Abelian group G and any prime divisor p of IGl there exists at least an element x, E G with ord(x,) = p. PROOF. Apply Proposition 1.4.10 and Proposition 1.4.8.
7. Prove that any finite field IFq satisfies the condition Cl (p ; a ) for any a E I$ and any p E IP with gcd(p, q - 1) = 1. 8. Let F be a field satisfying the condition Cl (n: a), with n 2 3. Prove that for any Abelian extension E/F which is not necessarily finite, the following assertions hold. E E i fi E F. (a) If n is odd, then @ E E ==+ E F. (b) If n = 4, then
9. Show that the implication " general.
*' +"
in Exercise 8 (b) does not hold in
10. With notation and hypotheses of Exercise 8, show that if n = 4 and fi E E then we may have & #i F. 11. Show that both statements (a) and (b) in Exercise 8 may fail for fields which do not satisfy the condition Cl (n; a). 12. (Bourbaki [40]). Let F be a field of characteristic p, let a E F*,let n E N* with p i n, and let N be a splitting field of the polynomial X7' - a. (a) Prove that the group Gal(N/F) is isomorphic t o a subgroup of the group I? of all matrices (: 7 ) with x E CT(Z,,) and Y E Zn. (b) Show that if n is a prime number, then the group Gal(N/F) is Abelian if and only if Cn E F or a E F". (c) Show that if F = Q, n E IP, and a # p,then Gal(N/K) is ' . isomorphic to the whole group I 13. With notation and hypotheses of Exercise 12, prove that if F satisfies the condition Cl (n; a ) and n # 4, then Gal(N/F) is Abelian if and only if a E F n . (Hint: Use Exercise 8.) 14. Cse Exercise 13 and Proposition 1.3.12 t o prove Exercise 12 (b). 15. Find for n 6 4 .
1.5. Exercises to Chapter 1
1. Show that the field Q does not satisfy the condition C1(4; -9), but satisfies the condition C1(4; -2). 2. Let F be a field satisfying the condition Cu(n;a). If 2 denotes the and d is the greatest coset of a in the quotient group F*/F*n divisor of n such that a E F*" then prove that
7 6 7 5 0 0 : ) Q]. 3. Find [Q( '

4. Let F be a field satisfying the condition Co(n;a ) , and let m = [ F ( f i ) : F]. Show that the map a : ID,, -+ Intermediate ( F ( *)IF)), a ( d ) = F ( %d ) , establishes an anti-isomorphism of lattices. 5 . Find all subfields of the field Q( fi)
D , ,. 16. Show that if G is a finite cyclic group of order n , then OG = I

17. Show that Lemma 1.4.3 may fail for a non Abelian finite group G. 18. Let G be a finite group of order n , and consider the canonical map Prove that the following assertions are equivalent. (a) G is a cyclic group.
, where n
E N* and a E
q.
6. Let p E IP and let F be a field such that p,(F) = (1). Prove that F satisfies the condition Cl (p ; a ) for any a E F*.
Chapter 1
(b) The map w c is injective. (c) The map wc: is a lattice isomorphism. Using Lemma 1.4.3 or Exercise 18, prove that any finite subgroup of the multiplicative group F* of any field F is cyclic. In particular, is cyclic. for any finite field IFq, the group (Darbi i53], [60]). Let F be a field of characteristic 0, and let f = X n - a E F [ X ] be an irreducible binomial with root u E 0. Let k = m a x i m E EDn ( u n l mE F(Cn) }. Prove that the degree of the splitting field of f over F is [ F(C,) : F] . n l k . Prove that the following statements hold for a finite extension E I F . (a) IGal(E/F)I [ E : F ] . (b) IGal(E1F)I = [ E : F] if and only if the extension E/F is Galois.
CHAPTER 2
<
KNESER EXTENSIONS
In this chapter we introduce and investigate two basic concepts of Cogalois Theory, that of G-radical extension and that of G-Kneser extension. Then we prove a criterion due t o Kneser [77], which characterizes Kneser extensions and plays a fundamental role in this monograph. Roughly speaking, a radical extension is an extension EIF such that E is obtained by adjoining t o the base field F an arbitrary set of "radicals" over F , i.e., of elements x E E such that xn = a E F for some n E N*.Such an x will be denoted by and will be called an n-th radical of x. Notice that this terminology is somewhat different from that used in Galois Theory (see e.g., Kaplansky [74], Karpilovsky [76], Lang [80]), but coincides for simple extensions. So, E/F is a radical extension when E = F ( R ) , where R is a set of radicals over F . Clearly, one can replace R by the subgroup (R) of the multiplicative group E* of E generated by F* and G = F* R. Thus, any radical extension E / F has the form E = F ( G ) , where G is a subgroup of E* containing F*,with G I F * a torsion group. Such an extension is called G-radical. A finite extension EIF is said t o be G-Kneser when it is G-radical and IG/F*I = [E : F ] .
1.6. Bibliographical comments to Chapter 1 - -
Section 1.1. The basic terminology and notation we use in this monograph are the standard ones, as those from Bourbaki [40], Kaplansky[74], Karpilovsky[76], and Lang [80]. Section 1.2. The review presented in this section is mainly based on Bourbaki [40], Kaplansky [74], Karpilovsky [76], a.nd Lang [80]. Section 1.3. Lemma 1.3.2 is due to Abel. The Vahlen-Capelli Criterion was first proved in 1895 by Vahlen [I071 for Q, and it was extended in 1897, 1898 by Capelli [43], [44] and in 1900 by Wendt [I121 t o arbitrary fields of characteristic zero. A summary of Capelli's results is given in his short Note [45] which appeared in 1901. The general result for fields of arbitrary characteristic was proved only in 1959 by Ridei [88, $1721. The proof of the Vahlen-Capelli Criterion we gave in this section basically follows the outline in Kaplansky [74], although there are several additions, simplifications, and improvements. The conditions Co(n; a ) and Cl (n; a) applied for fields were introduced by Albu [3]under different names. Section 1.4. Most of the results in this section are essentially folklore.
2.1. G-Radical and G-Kneser extensions In this section we introduce and investigate two basic notions of this monograph, that of G-radical extension and that of G-Kneser extension. For any extension EIF we shall use throughout this monograph the following notation. T ( E / F ) := { x E E* I xrLE F* for some n E
W ).
&
Clearly F* T ( E / F ) , so it rnakes sense t o consider the quotient group T ( E / F ) / F * . Remember that for any x E E we denote by 2 the coset x F * of x rnodulo F* in the group E*/F*.
53
<
54
Chapter 2
2.1 G-Radical arid G-Kneser ext.erisior~s
55
Observe that the quotient group T ( E / F ) / F * is precisely the torsion This group, playing a.major group t ( E * / F * ) of the quotient group E*/F*. role in this monograph, is somewhat dual t o the Galois group of E I F , which explains the terminology below. DEFINITIONS 2.1.1. The Cogalois group of a n arbitrary field extension E I F , denoted by Cog ( E I F ) , is the quotient group T ( E / F ) / F * . A finite extension EIF is said t o be a Cogalois extension if E = 0 F ( T ( E / F ) ) and [Cog(E/F)I = [ E : F ] . In the next chapter we will characterize and investigate Cogalois extensions and will compute the Cogalois group of some extensions. Observe that for every element in x E T ( E / F ) there exists an n E N* and is called such that x n = a E F. Such an x is usually denoted by an n-th radical of a. Thus, T ( E / F ) is precisely the set of all "radicals" belonging t o E of elements of F . This observation suggests t o define a radical extension as being an extension EIF such that E is obtained by adjoining to the base field F an arbitrary set of "radicals" over F. More precisely, we have the following definition. DEFINITION 2.1.2. A n extension EIF is said t o be radical if there exists A & T ( E / F ) such that E = F ( A ) . W e say that E/F is a simple radical extension if there exists a n a E T ( E / F ) such that E = F ( a ) . 0 Clearly, one can replace the set A in Definition 2.1.2 by the subgroup G = F * ( A ) of the multiplicative group E* of E generated by F* and A. So, any radical extension EIF has the form E = F ( G ) , where G is a subgroup of E* containing F * , with G I F * a torsion group. For such extensions we will use a special name. DEFINITION 2.1.3. Let E/F be a n extension and let G be a group. T h e n , EIF i s said t o be a G-radical extension if F* G T(E/F) and E = F ( G ) . 0
EXAMPLES 2.1.5. The examples below show that the Cogalois group of a finite radical extension could be infinite.
(1) Consider the extension IF2 (X)/IF2(X2) of degree 2. We claim that Cog (F2iX)/IF2 ( X 2 ) ) = F2 (X)*/F2(X2)* is a countably infinite group isomorphic t o a countably infinite direct sum of copies of Z2. Since each element of this group has order 2, our claim will follow once we prove that this group is infinite. To do that, observe that there exist infinitely many irreducible polynomials in the ring IF2 [XI, and for any distinct irreducible polynomials p, , pp in F2 [XI, their cosets p;' , fi in Cog (IF2(X)/F2 ( X 2 ) ) are distinct. Indeed, assume that p? = $2. Then, there exist f , g E IF2 [XI ? It folsuch that gcd(f, g) = 1 and p, /p2 = f 2 / g 2 , that is, plg2 = pa f lows that both polynomials f and g are nonconstant, so, decomposing them into irreducible polynomials, and using the fact that IF2 [XI is an UFD, we deduce that necessarily pl = pp, which proves our claim.
<
(2) Similar arguments show that IF, (X)/IF,(Xp), where p > 0 is any prime number, is an extension of degree p with Cogalois group isomorphic 0 t o a countably infinite direct sum of copies of the cyclic group Z,.
LEMMA 2.1.6. Let E/F be a G-radical extension, which is not necessarily finite. T h e n any set of representatives of the factor group G I F * is a set of generators of the F-vector space E. I n particular, one has
IG/F*I
[E : F].
PROOF.Let T = { t i 1 i E I) be a set of representatives of the factor group G I F * . Then G / F * = { G J i E I). For each z E E = F ( G ) = F[G] there exist XI,. . . ,As E F and g l , . . . ,gs E G such that x =
<
<
&
gk
E G / F * , so g^l = t i , , . . . ,gs = t i * , where i l l . . . , i s E I. It follows that = pktik for some pk E F * . Consequently

h
k= 1
C X k g k . But
REMARKS 2.1.4. (1) Clearly, a G-radical extension E/F is also GIradical for any G' with G G' T ( E / F ) .
< <
(2) Let F g K g E be a tower of fields. If EIF is a radical extension, then clearly so is also the extension E I K . We will see in Section 3.2 that, in general, the extension K I F is not radical, and if K I F and E I K are both radical extensions, then EIF is not necessarily so.
(3) Any radical extension is clearly algebraic. This implies that if E/F 0 is a G-radical extension, then E = F ( G ) = F[G].
with vk = Xkpk E F . Thus T is a set of generators of the F-vector space E. 0
COROIJARY 2.1.7. Let E/F be a finite G-radical extension. T h e n there exists a subgroup H of G such that H I F * is a finite group and E / F is H -radical.
56
Chapter 2
PROOF. Let S be a fixed set of representatives of the factor group G I F * . By Lemma 2.1.6, S is a set of generators of the F-vector space E . Now extract from S a finite basis B = {bl,. . . ,b,) of this vector space, and set H = F * ( B ) . Since the Abelian group HIF* is generated by the finite set {bl, . . . ,b,) consisting of elements of finite order, we deduce that
e
A
that ;x # ;x for each j l # j2 in J . If { x l , . . . ,x,,) is a finite subset of M, then we have G = t i , , . . . ,x,, = t i n , with i l , . . . ,i, E I. It follow~ that X I = Xlti,,. . . ,X, = Xr1ti, for some X I ? . . .,A,, E F * . The set {ti,, . . . , t i n ) is clearly linearly independent over F , since it is a subset of . ,x,) is linearly independent over F , so the basis T. It follows that { x 1 7 . . M = { x j I j E J) is linearly independent over F .
h
r. is a finite group of order n l . . . . n,, where ni = ord(bi) 1 i Since E = F[B] = F[H] = F ( H ) , we conclude that the extension EIF is 0 H-radical. The next result works for G-radical extensions which are not necessarily finite, and will be also used in Part 2 for infinite extensions.
<
< <
(5) (7): Let H be a subgroup of G such that F* < H and HIF* is a finite group, and let S be a fixed set of representatives of the factor group H I F * . The set S is a subset of G consisting of elements with distinct classes, and by assumption, it follows that S is linearly independent over F . This shows that IH/F*I < [ F ( H ) : F]. (7) ==+ (2): Let T = { ti 1 i E I) be a set of representatives of the factor group G / F * and let i t l , . . . ,t,,) be a finite subset of T. Consider the subgroup ( 6 , . . . t,,) of G I F * generated by 16,. . . , ). This subhas finite order in G I F * . Clearly group is finite since each 6 , . . . , ( 6 , .. . , r$ ) = H I F * for some H G with F* H . By assumption, we have IHIF*I [ F ( H ) : F], and by Lemma 2.1.6, applied t o the H-radical extension F ( H ) / F , the opposite inequality also holds. Thus ) H / F *I = [ F ( H ) : F 1, hence any set of representatives of the finite group HIF* is a vector space basis of F ( H ) over F . Since {tl , . . . ,t,) is a subset of such a set of representatives, it follows that it is linearly independent over F . This proves that T is linearly independent over F . J T
PROPOSITION 2.1.8. The following assertions are equivalent for an arbitrary G-radical extension E I F . (1) There exists a set of representatives of the factor group G I F * which is linearly independent over F . (2) Every set of representatives of G / F * is linearly independent over F . (3) Every set of representatives of G I F * is a vector space basis of E over F . (4) There exists a set of representatives of G I F * which is a vector space basis of E over F . (5) Every subset of G consisted of elements having distinct cosets in the group G / F * is linearly independent over F . for each (6) Every finite subset {xl, . . . ,x,) C G such that ifi # i, j E (1,.. . ,n), i # j, is linearly independent over F . (7) For every subgroup H of G such that F* H and HIF* is a finite group, IH/F*I [ F ( H ) : F ] .
cl
<
<
<
<
<
As we showed in Lemma 2.1.6, for any G-radical extension EIF we have the inequality IG/F*I 2 [E : F]. The finite extensions for which we have the opposite inequality deserve a special name. DEFINITIONS 2.1.9. A finite extension EIF is said to be G-Kneser if it is a G-radical extension such that IG/F*I [E : F ] . The extension EIF 0 is called Kneser if it is G-Kneser for some group G.
PROOF. (1) ===+ (2): Suppose that there exists a set of representatives T of G / F * which is linearly independent over F . Let S be a n arbitrary set of representatives of G / F * . Clearly, for each s E S there exists t E T with j: = i.e., s = At for some X E F * . This implies t h a t S is linearly independent over F because T is so.
(2) ===+ (3) follows from Lemma 2.1.6. (3)
<
* (4), (4) a ( I ) , and (5) u (6) are obvious.
(3) a (5): Let T = { ti I i E I ) be a set of representatives of G I F * . Then T is a basis for E over F . Let M = { x j I j E J) C G be such
COROLLARY 2.1.10. The following assertions are equivalent for a finite G-radical extension EIF. (1) E / F is G-Kneser. (2) I G / F * I = [ E : F ] . (3) Every set of representatives of G I F * is a vector space basis of E over F . (4) There exists a set of representatives of G I F * which is linearly independent over F.
Chapter 2
2.1 G-Radical and G-Kneser extensioris
59
(5) Every finite subset {xl , . . . , x,) E G such that Fi # f for each i, j E (1,. . . , n ) , i # j , is linearly independent over F. (6) The extension F ( H ) / F is H-Kneser for every H , F* H G.
< <
which shows that the extension Q(4-)/Q is simultaneously G-Kneser and H-Kneser. However, G # H since fi E H \ G. 0 The uniqueness of the group G for the so called G- Cogalois extensions, which are separable G-Kneser extensions possessing a certain inheritance property, will be discussed in Section 4.4. REMARKS 2.1.14. (1) A subextension of a Kneser extension is not necessarily Kneser. Indeed, we will see in Section 5.3 that the extension Q( J ~ ) / Q is not radical, and so, it is not Kneser. On the other hand, it is easily seen that Q( J2 + G Q(Ga), and by Exercise 16, Q(C16)/Q is a p (cis)-Kneser extension.
PROOF. Apply Lemma 2.1.6 and Proposition 2.1.8.
PROPOSITION 2.1.11. Let EIF be a finite G-Kneser extension. Then, the extension F ( H ) / F is H-Kneser and F ( H ) n G = H for every H with F* H G.
< <
PROOF.The extension F ( H ) / F is H-Kneser by Corollary 2.1.10. We are going to prove the equality F ( H ) n G = H . The inclusion H G F ( H ) n G is obvious. Now let x E F ( H )n G. Then x = Xlhl . . . + Ash, for some Xi E F, hi E H . We can assume that the classes h l , . . . , h, are mutually distinct in the group H / F * and so, also in the group G I F * . Since x, h l , . . . ,h, are linearly dependent over F, it follows by Corollary 2.1.10 that there exist at least two equal cosets among x, h l , . . . , h,. This is possible only if 2 = hi for some i E 11,. . . ,s), i.e., x = Ahi with X E F * , and so x E H . This proves the-inc1usi;n F ( H ) r7 G H.
h
a)
(2) Also, we will see in Section 3.2 that the property of an extension being Kneser is, in general, not transitive. 0 EXAMPLES 2.1.15. (1) The extension Q(J q ) / Q is Q? but it is not p (C3)-Kneser. Note that Q(J-3) = Q(C3 ). Indeed, since
G2 = -3
Q ( n ) / Q = (7,
f i )hence ,
E Q , it follows that ( a ) 2 = 7, that is,
(q) -Knew
COROLI~ARY 2.1.12. Let E / F be a finite extension, which is simultaneously G-Kneser and H-Kneser. If H G, then H = G.
<
I Q ( a ) / V On the other hand, since C3, [:, 3 ) ={ ,,
c/c3 # Q+ and [i = 1 E Q+,we have

IQ+ (C)/Q+I = 3 > [Q(G) : Q] = 2.
1 =2=
: Q].
PROOF. We have G = E n G = F ( H ) n G = H according t o Proposi[? tion 2.1.11.
} and
EXAMPLE 2.1.13. The next example shows the non-uniqueness of G for a given G-Kneser extension. Consider the extension Q ( 9 / q ) / Q , where is one of the complex roots of the irreducible polynomial X 4 + 9 E Q[X], say &(l + 412. Since ( m ) 2 = 3i, we have i = f ( m ) ' E Q ( f q ) , and so, fi = 2 m / ( 1 + i) E Q ( 9 / 9 ) . Now consider the groups G = p and H = p ( i , & ). Then
(2) A Cogalois extension E/F is nothing else than a T(E/F)-Kneser extension. These extensions will be discussed in the next chapter. (3) Any Cogalois extension is a Kneser extension as mentioned in the previous example. The converse of this statement does not hold: the extension Q(J ~ ) / Qis Q ( a ) - ~ n e s e by r (I), but it is not Cogalois. Indeed, we have shown a t (1) that the group Q (b)/p has order 3. Since this group is a subgroup of the group Cog J-)/Q), it follows that I C ~ R ( Q ( J - ~ ) / Q2 ) )3~> [ Q ( J - ~ ) : Q] = 2. This mearls that the extension Q(-)/Q is not Cogalois. We will show in Section 3.3 that
(m)
(a
and
But Q(J;/-9)
= Q(i, JG),
hence
G ) G, c ~ ~ ( Q ( J - ~ ) /= Q) that is, the Cogalois group of this extension is a cyclic group of order 6. 0
(a,
60
Chapter 2
2.2 'I'he Krieser Criterion
61
2.2. The Kneser Criterion In this section we prove a nice result, due to Kneser [77], which characterizes finite separable G-Kneser extensions E/F according t o whether or not certain roots of unity belonging to G are in F .
i) Char(F) = 2. Then [j = 1, hence
ii) Char(F) # 2. Since (1 + C4)4 = -4 E F* it follows that the element 1 + c4 E G / F * has order 1 or 2 or 4.
= 1, then 1 5 4 E F * , so C E F . If o r d ( l + If ord(1 (4) = 2, then (1 +~4)" F*,so 2C4 E F*;since Char(F) #2 . it follows that C4 E F. If ord(1 t c4) = 4, then the elements
,
THEOREM 2.2.1 (THE KNESERCRITERION). The following assertions are equivalent for a separable G-radical extension E/F with finite G/F*. (1) E/F is a G-Kneser extension. (2) . . For every odd prime p, 5, E G ==+ Cp E F , and 1 i C4 E G ==+ C4 E F . (3) pp(G) = p p ( F ) for every odd prime p, and 1 L C E G =+ C4 E F .
PROOF. First of all, note that E/F is necessarily a finite extension, since E = F ( G ) and G I F * is a finite group. (2) W (3): Assume that pp(G) = p p ( F ) for every odd primep. Then 6E F E G 6 E F . Conversely, assume that &, E G clearly for every odd prime p, and let g E pp(G). Then gp = 1, hence the order ord(g) of g in G is either 1 or p. If ord(g) = 1 then g = 1 E F , as desired. If ord(g) = p then g generates the subgroup (g) = (C,) of order p of G. Thus C, E (g) G, hence Cp E F by our assumption. It follows that g E (Cp) F*, as desired.
c) + -
C4 = 1 E F .
are distinct in G / F * . By Corollary 2.1.10, the elements 1, I C4, (1 C4)2, (1 C4)3 are linearly independent over F , which is a contradiction because (1 + C4)2= 2C4, 2 E F*,and
*
<
This shows that we cannot have ord(1 C4) = 4, so necessarily we must have ord(1 C4) E {1,2), and then C4 E F, as shown above.
h -
(2) ==+ (1): Suppose that the conditions in (2) are satisfied, and show that IGIF*I ,< [E : F ] . Of course, we can assume that IG/F* I 3 2. To do that, it is sufficient t o prove the i~nplication only when the group G / F * is a pgroup, that is, G / F * has order a power of a prime number p. Indeed, if IG/F* I = p:' - . . . . p k , with pi mutually distinct prime 1 < i < r , let Hi/F* be a pi-Sylow subgroup of numbers and ti E N*, i = 1 , . . . , r. Assu~ning that the result holds for G / F * , i.e., IHi/F*I pgroups, we deduce that pfi = [ F ( H i ) : F ] . But [ F ( H i ) : F ] divides E : F j, hence p:i divides [ E : F j, and so, divides [ E : F ] . It follows that IG/F* ( divides [ E : F 1. Thus
<
(1) ==+ (2): Let us specify that the condition 1f 6 E G means that either I C4 E G or 1 - E G. 1- C4 E G. Indeed, this is clear whenever Observe that 1+ (4 E G Char(F) = 2. If Char(F) # 2 and 1 + C4 E G, then (1 + C4)2 = 2 6 E G. = -1 + (4 E G. and then hence 5 4 E G since 2 E F* G. Thus C(1 + 6) 1 - C4 = (-I)(-1 + (4 .E G.
c4
Suppose that the extension E/F is G-Kneser and 6 E G, where p is an odd prime. Then = I , hence o r d ( c ) in the group G I F* divides p, that is, o r d ( c ) E {l,p}. -.p-1 would be If we would have ord(C,) = p, then the covets 1, . . . , 6 distinct in G / F * , hence the subset {I, Cp, . . . , CpP-' } of G would be linearly independent over F by Corollary 2.1.10, which would contradict the fact that 1 + Cp + . . . + Ci-l = 0. Consequently o r d ( 6 ) = 1, and this implies that Cp E F * , as desired.
cP ,.
A
n:=,
- G, -
(G/F*I [ E : F ] . Consequently, without loss of generality we can assume that IG/F*I = pt, with p E P and t E N*. There exists then an ascending chain of subgroups of G such that JHs/H,-I I = p, s = 1,.. . ,t. We are going t o prow by induction on s , 0 s t, the following two statements.
<
< <
If 1 +
E G, we investigate two cases:
(4
[ F ( H s ) : F(Hs-I )I = P.
Chapter 2
2.2 The Krieser Criterion
63
(23,)
If p > 2, and c E F ( H s ) with cp E Hs, then c E Hs. If p = 2, and c E F (H,) with c2 E H,, and either c E G or C4 $! F(H,), then c E H y.
Since 1 q exist u , v E
<
< p, we deduce that p and q are relatively prime, hence there i Z such that up + vq = 1. Consequently,
Observe that ( a o ) is obvious and (Ao) has no meaning. Let s 2 1. implies (A,). Let a E H, be such that the First, we show that coset iZ = U H , - ~ is a generator of the cyclic group Hs/Hs-l of order p. Then a H , \ H,-l, H, = HSwl(a), and ap E H,-l. Clearly, a is a root of the polynomial f = XP -up E F(H,-') [XI. We claim that this polynomial is irreducible over F(H,-'), and then it will be the minimal polynomial Min(a, F (HSu1))of a over F(HS-'), which will imply that [ F ( H , ) : F(H,-l)] = p, as desired. If f would be reducible, then, by Lemma 1.3.2, it would have a root b E F(Hs-1 ). Then bp = ap E H,-1. If p > 2, then we have b E Hs-l by (B,-l). Since b = aCp, a E Hs, and b E HSw1,we have Cp = ba-' E H, G. In view of (2), this E HS-1, contradicting the would imply that Cp E F and so, a = bC;' choice of a. " b2, hence a = fb, and so, b E H, G. As before, If p = 2, then a by (a,-'), we have b E H,-l, hence a E H,-l, which is a contradiction. Thus, for any prime p, the polynomial XP - ap is irreducible over F(H,-l), i.e., (A,) holds.
Since up, N(c)d-' E F(HS-'), we deduce that ap is a p-th power of an element from F(H,-l), which contradicts the irreducibility of the polynomial f = XP - aP E F(H,-l)[X]. Thus, we must have q = 0, and then cp = d E HSv1. For simplicity, set K = F(H,-I), and observe that F ( H s ) = K ( a ) since Hs = H,-1 (a). Let L be the normal closure (contained in Cl) of K ( a ) / K . Since the extension E / F is separable, we deduce that L / K is a finite Galois extension, hence a E L \ K implies that there exists cp E Gal ( L I K ) such that cp(a) # a. Since aP E H,-l, we have cp(aP) = ap = cp(a)P, hence cp(a) = aC, Similarly, CP E Hy-I implies that cp(cP) = cP = cp(c)P, hence cp(c) = cC7 for some 0 m p - 1. Then
<
< <
<
We claim that a-'"c E F(Hs-1) = K. Indeed, set w = a-'"c. Min(a, K ) = XP - up, there exist Xi E K , 0 i p - 1, such that
< <
Since
and (A,) together imply (a,). Now, we are going t o prove that We will examine separately the cases when p is odd or even.
Case 1: If p > 2, let c E F(H,) be such that cp E H,. Then Ep = iiq, with 0 q p - 1, hence CP = aQd for some d E Hs-l. We
Since p ( a ) = Cpa, we deduce that
want t o show that c E H,. Assume first that q > 0, and denote by N the norm of F (H,) over F (H,-'). Since Min(a, F (H,-l)) = XP - up, we have N ( a ) = (-1)P-lap. Taking the norm N in both sides of the equality
< <
But [ K(Cp) : K ] p - 1 and [ K ( a ) : K ] = p, hence these degrees are relatively prime. Then
<
and so, we obtain In particular, it follows that the set {I, a , . . . ,up-') is linearly independent over K(Cp). Therefore, Xi = for all i, 0 i p - 1. If we assume that w $! K , then there would exist 1 j p - 1 such that Xj # 0, which would imply that q = 1. But this contradicts the fact that CP # 1.
XiCi
< <
But p has been supposed t o be odd, hence ( t ) becomes
< <
64
Chapter 2
2.3 Exercises to Chapter 2
65
Since Thus, we have proved that w = a-"c E K = F(H,-,). (U-"~C)~' = (UP)-~'~CP E H S p l , the staternent (B,-l) implies that aa-,c E Hs- , hence c E H,.
Case 2: If p = 2, let c E F(H,) be such that 6 ' E H,, and either c E G or (;1 # F(H,). We have t o prove that c E H,. As in Case 1, we have, c2 = a4d for some 0 6 q < 2 and d E H,-l. If q = 1 then c2 = ad, hence (t), which also holds for p = 2, implies that -a" z2 for some z E F(H,-l), and then a = kc4z. This implies # F(H,-l), F(H,) = F(Hs-1)(c4), and c2 = ad = f c 4 z d . It that follows that c = x C4y for some x, y E F(Hs-l). Thus
g E H . Since g was an arbitrary element of G, we deduce that H = G. This proves that G / F * = H I F * is a finite group.
c4
xZ = y2, i.e., x = f y, hence c = (1 ZIZ c4)x. Observe that necessarily = 1, and then a = z E Char(F) # 2, for otherwise, it would follow that F(H,- ) , which is a contradiction. On the other hand, c2 E H, and IH,/H,-ll = 2, hence c4 = ( c ~ E ) ~ H,-'. Then x4 = (-c4) . 4-' E H,-l. By it follows that x2 E H,- 1 , and applying (23,- ) again, we deduce that x E H,-l. Then 1zk = cx-' E G, and by our assumption in (2) we deduce that E F F F(HSFI ), which is a contradiction. Hence q is necessarily 0, and then c2 = d E HS-l. Kow we will proceed as for p > 2 by taking an automorphism p E Gal ( F ( H s ) / F ( H s - I ) ) such that p ( a ) = -a. We deduce that ip(ajc) = a J c for a suitable integer j, and a j c E F(HSe1). Since a2jc2 E H,-1 , the statement (23,-1) implies that a j c E H,-I , and so, c E Hs.
SO
c4
(2) The next example shows that the separability condition cannot be dropped from the Kneser Criterion. Take the extension E/F considered in Example 2.1.5 ( I ) , that is F = IF2 ( X 2 ) and E = F2 ( X ) . We have seen that Cog ( E I F ) = T ( E / F ) / F * = E*/F*is an infinite group. Observe that E/F is T(E/F)-radical since E = F ( X ) and X 2 E F . Further, the condition (2) from the Kneser Criterion is satisfied for G = T ( E / F ) since p n ( E ) = (1) for every n E W . Indeed, let n E W and f E E be such that f n = 1. This means that the element ,f E E is an integral element over IF2 [XI. Since IF2 [XI is an integrally closed domain, we deduce that necessarily f E F2 [XI, and moreover, f is a unit of the domain 1F.L [XI. Then f is a nonzero constant polynomial in IF2 [XI, that is, f = 1, as desired. However, the finite extension E/F is not G-Kneser, since
c4
c4
(3) We do not know whether or not the separability condition in the Kneser Criterion can be weakened. 0
-4pplications of the Kneser Criterion to algebraic number fields will be given in Chapter 9.
This completes the inductive proof of (A,) and (23,). From (A,) we now deduce that
[ E : F] = [ F ( H t ) : F ( H t - l ) ] . . . . . [ F ( H l ) : F ( H o ) ] = pt.
which proves the implication (2) (I), and we are done.
1. Prove that pn ( F( X I , . . . ,X,)) m,nEN*.
= pn ( F ) for any field F and any
RXMARKS2.2.2. (1) The condition "G/F* is a finite group" in the statement of the Kneser Criterion can be replaced by the condition " E I F is a finite extension". Indeed, since EIF is a finite G-radical extension, E = F ( H ) for a suitable group H , with F* H G and HIF* finite, by Corollary 2.1.7. Then E/F is H-Kneser by the Kneser Criterion applied t o the H-radical extension E/F with finite H I F * . Let g E G, and set H ' = H(g). Then EIF is also HI-Kneser, again by the Kneser Criterion, m d H 1 / F * is a finite group. By Corollary 2.1.12, we have H = H', hence
2. Let p be any positive prime number, and let n be any positive F ' , ( X I , . . . ,Xn)/IFp ( X r , . . . ,X,P) integer. Prove that the extension I is an extension of degree p7" with Cogalois group isomorphic t o a countably infinite direct sum of copies of the cyclic group Z,.
3. Show that Cog(F4/IF2) 2
< <
K.
IFqn
4. Prove that for any extension
/IFq of finite fields, one has
Chapter 2
2.4 Bibliographical comments t o C h a p t e r 2
67
5. (Greither and Harrison [63]). Let E = Q({ C2= I n E W )) and F = E n R. Prove that E = F ( i ) , [ E : F] = 2, and Cog ( E / F ) is a countably infinite group.
6. (Dummit [55]). Let p be an odd prime, let E = Q(Cp), and let F = E n R . Showthat [ E :F] = 2 and ICog(E/F)I = 2p.
19. (Gay and Vdez [61]). Let F be any field, let u E R be a root of a binornial X n - a E F[X], and let m = ord(6) in the group R*/F*, where R in an algebraically closed overfield of F. If gcd(n, e ( F ) ) = 1 and C2, 61 F ( u ) \ F for every prime divisor p of n , then prove that [ F ( u ) : F] = m. 20. (Risman). Let E / F be a separable extension, and let u E T ( E / F ) . If m = ord(6) in the group E * / F * , then prove that m = nt, where gcd(n, t ) = 1, n 1 [ F ( u ) : F ] , and hp # F ( u ) \ F for every prime divisor p of t. (Hint: See [29] and [61].)
7. (May [81]). Let F be a field, and let p be a prime different from the characteristic of F . Let u E R with UP E F \ FP, and let E = F ( u ) . If p = 2, further assume that E # F(C4). Prove that Cog(E/F) = ((u)t(E*)F*)/F*. 8. (May [81]). Let F = Q , p = 2, u = C4 = i, and E = F ( u ) . Show that 1 i E Cog(E/F) \ ((u)t(E*)F*)/F*.
21. (Schznzel [92]). Let F be any field. Prove that the Galois group of the splitting field of a binomial X n - a E F [ X ] is Abelian if and only if awn E F n , where wn = Ipn(F)I. 22. (Schinzel [92]). Let F be any field, and let f = X n - a E F [ X ] be an irreducible binomial such that the Galois group G of the splitting field of f is Abelian. Prove that if 4 1 n and C4 @ F, then G EZ Z 2 x Z r l p , otherwise G is cyclic.
23. (Halter-Koch [65]). Let E/F be a finite separable extension, let n E W be such that gcd(n,e(F)) = 1, and let G be a group such that F* G 6 E* and GrLC F . Prove the following facts. : F * n )I [ E : F ] (a) (GTL (b) G / F * is a finite group.
9. (Brandis [41]). Let F be an infinite field, and let E be an overfield of F with E # F . Prove that the group E*/F*is not finitely generated.
10. Let E = & ( X ) and F = IF2 (X". Show that the extension E/F is simultaneously F*( X )-Kneser and F*( X + 1)-Kneser, but the ( X ) and F*( X + 1) are distinct. groups F* 11. Prove that any quadratic extension E / F , with Char(F) Kneser . 12. Prove that the extension IF4 /IF2 is not Kneser. 13. Investigate whether or not the quotient extension of a Kneser extension is also Kneser. 14. Is the compositum of two Kneser (resp. radical) extensions again a Kneser (resp. radical) extension? 15. Give an example of a radical extension which is not a Kneser extension. 16. Let n E W . Prove that Q(Cn)/Q is a and only if n = 2' for some r 6 N. (Cn)-Kneser extension if
# 2 is
2.4. Bibliographical comments to Chapter 2

Section 2.1. The concept of radical extension is rather basic and wellknown in Galois Theory. However, our terminology is somewhat different from that used in Galois Theory (see e.g., Kaplansky [74], Karpilovsky [76], Lang LBO]), but coincides for simple extensions. Note that radical extensions are called coseparable by Greither and Harrison [63]. The concept of Gradical extension is due to Albu and Nicolae [19]. Radical extensions were studied among others by Acosta de Oroxco and VClex [I], Barrera-Mora and VClez [32], Gay and V&lez [GO], Halter-Koch [65], [66], Norris and VClez [86] [I091, Velez [l 081, [l101. The torsion group t ( E * / F * ) of an extension E/F was intensively investigated by Acosta de Orozco and VClez [2], Gay and VClex [61]. To the best of our knowledge, the name of Cogalois group of E/F for the group t ( E * / F * ) appeared for the first time in the literature in the fundamental
17. For what n E N* is Q(C,, ) / Q a Kneser extension? 18. Let F be an arbitrary field, let u E R be any root of an irreducible binomial X n - a E F [ X ] , arid let m = ord(6) in the group R*/ F * , where R in an algebraically closed overfield of F . Prove that the extension F ( u ) / F is F*(u)-Kneser.
68
Chapter 2
paper [63] of Greither and Harrison. The term of "coGalois group" was also used by Enochs, Rozas, and Oyonarte [56], [57], but with a completely different meaning, involving the concept of F-cover of a module. The concept of Kneser extension was introduced by Albu and Kicolae P91. This section basically follows the outline in Albu and Nicolae [19] and Albu and Tena [25], although there are several extensions, simplifications, and i~nprovements.
CHAPTER 3
Section 2.2. The Kneser Criterion which, of course, is due t o Kneser [77] was published in 1975. Actually, Kneser proved only the difficult implica(1) in Theorem 2.2.1. The other implication (1) tion (2) (2) in Theorem 2.2.1 is mentioned by Schinzel [91], Gay and Vdlez [61], and it was explicitly proved for the first time in 1982 by Schinael [93]. The very thorough proof of the Kneser Criterion presented in this section details the original ideas from the very concise Kneser's paper [77], and follows the more accurate proof given in Schinzel [93] (see also Schinzel [94]), which corrects a little gap in the original proof of Kneser [77]. The Kneser Criterion is not only a basic tool in the whole Cogalois Theory, but it has nice applications t o classical Algebraic Number Theory (see Chapter 9) and also to Grobner bases (see Becker, Grobe, and Niermann P4I ).
COGALOIS EXTENSIONS
The aim of this chapter is t o investigate the Cogalois extensions, introduced into the literature in 1986 by Greither and Harrison [63]. Using the concept of Kneser extension, a finite extension E / F is a Cogalois extension precisely when it is T(E/F)-Kneser. We defined in Section 2.1 the Cogalois group of an extension E/F as being the group T ( E / F ) / F * and we denoted it by Cog(E/F). Thus, a finite radical extension E/F is Cogalois if and only if 1 Cog ( E lF ) I = [E : F]. Using the Kneser Criterion we provide a short proof of the GreitherHarrison Criterion characterizing Cogalois extensions. We present then the Gay- Vdea Criterion, which is an equivalent form of the Greither-Harrison Criterion. Some simple properties of Cogalois extensions are also given. Fi. , nally, we calculate explicitly the Cogalois group of any quadra.tic extension of Q.
3.1. The Greither-Harrison Criterion

The aim of this section is to state and prove a criterion characterizing Cogalois extensions in terms of purity, due t o Greither and Harrison [63]. Our simple proof is based on the Kneser Criterion. We also relate the Krieser Criterion t o the Gay-Vdlea Criterion. R.ecal1 that throughout this monograph we shall use the following notation:
= {
p W~ l p prime),
P = {2)) u {4h Recall also that R is a fixed algebraically closed field containing the fixed base field F as a subfield; any considered overfield of F is supposed t o be a subfield of R. For any n E W , C , will denote a primitive n-th root of unity over F , i.e., a generator of the cyclic group p,(R).
7
P\
69
I
70
Chapter 3
3.1 T h e Greither-Harrison Criterion
71
A fundamental concept in the theory of radical extensions is that of purity, which is somewhat related to that used in Group Theory. Recall that a subgroup H of an Abelian multiplicative group G is called pure if G7' n H = Hn for every n E N*. DEFINITION 3.1.1. An extension E/F is said to be pure if p p ( E ) C F 0 for every p E P . LEMMA 3.1.2. The following assertions are equivalent for an extension E/F. (1) E/F is pure. (2) P,(E) = P,(F) for every P E P. (3) Cp E E Cp E F for every p E P . (4) 52, $ E \ F for every p E P.
-Gp E E \ F , that is 6 E E \ F , which again contradicts (3). This proves the implication (3) =j (4).
Now assume that (4) holds. Since C4 $ E \ F by hypothesis, it is obvious that C4 E E C4 E F . Let p be an odd prime, and assume that Cp E E . Then fC2, E E in view of the considerations above. If we would have C, $ F , this would imply that f 6,$ F , i.e., C2, E E \ F, which contradicts (4). This proves the implication (4) ==+(3). 0 REMARK 3.1.3. Consider the extension EIF with E = Q ( 6 ) and F = Q, where p is an odd prime. Then Cp2 $ E , for otherwise, it would follow that Q(Cp2 ) = Q((, ) , hence
PROOF. ( I )
(2) and (2)
* (3) are obvious.
(3) (1): Assume that (3) holds, and let C E p, (E). Then Cp = 1, hence the order ord(C) of C in E* is a divisor of p. If p is an odd prime, ord(C) is either 1 or p. If ord(C) = 1, then C = 1 E F , as desired. If ord(C) = p, then C generates the subgroup (C) = (C,) of order p of R*. Thus C, E (C) E * , hence C, E F by our assumption. It follows that C E (C,) F * , as desired. Kow assume that p = 4. Then C4 = 1, hence ord(<) E {1,2,4). If ord(<) = 1 then C = 1 E F , and if ord(C) = 2 then C = -1 E F . If ord(() = 4, then as in the case of an odd prime, we have C E (C4) F * . Thus, E/F is a pure extension.
which is a contradiction. Hence Cp2 E E j Cp2 E F holds v a c ~ o ~ s l y . On the other hand, (C,) = pp2 ( E ) # ppp2 ( F ) = {I). This shows that 0 implication (3) j (2) in Lemma 3.1.2 may fail when p is not in P. EXAMPLES 3.1.4. (1) A field F is said to be a field with few n-th roots (F) {-1,l); when this holds for every of unity, where n E K , if p,& n E N*, i.e., if p ( F ) 5 {-I, I}, then F is said t o be a field with few roots of unity. Clearly, any extension E/F with E any field with few roots of unity is a pure extension, and any subfield of R is a field with few roots of unity. In particular, any extension E / F , where E is any subfield of W is pure. More generally, any extension E I F , with p,(E) F for every n E K is obviously pure. Thus, for any field F and any m E K ,the extension F(X1, . . . ,X,)/ F is pure, by Exercise 1, Chapter 2. In particular, the extension F2 (X)/lF2 ( X 2 ) is pure.
<
<
<
(3) W (4): First of all, observe that if Char(F) = 2, then clearly C2, is a primitive p-th root of unity for every odd prime p. Indeed. since Char(F) = 2, the equality = 1 implies that C; , = 1, hence the order of C2, in R* is a divisor of p. Since clearly (2p # 1, we deduce that ord(C2,) = p, that is, C2, is a primitive p t h root of unity. If Char(F) # 2, we claim that -C2, is a primitive p t h root of unity for every odd prime p. Indeed, we have ( C : , ) ' = 1, hence $ = -1, and then (-C2,)P = (-1)PC;, = (-I)(-1) = 1. If Char(F) = p, then -C2, = Cp = 1. If Char(F) # p, then -C2, # 1 since ord(CiLp)= 2p > 2, hence ord(-C2,) = p. This proves our claim. Assume that (3) holds and C2, E E\F for some p E IP. Then necessarily (4 E F . If Char(F) = 2 then we p is odd since, by (3), C4 E E have Cp E E \ F , which contradicts (3). If Char(F) # 2, then we have
(4
= 1 and
c ;:.
(2) A quadratic extension Q ( J ~ ) / Q where d is a square-free integer is pure if and only if d # -1, -3 (see Corollary 3.3.3). 0
PROPOSIT 3.1.5. I ~ NLet F C K E be a tower of fields. Then E/F is pure if and only if both K / F and E I K are pure.
PROOF. Clearly pp ( F ) C pp(K) p, ( E ) for every p E P , hence
which proves the proposition.
LEMMA 3.1.6. Let EIF be a finite separable G-radical extension. If E/F is pure, then E/F is G-Kneser, and G = T ( E / F ) .
72
Chapter 3
3.1 T h e Greither-Harrisorl Criterior~

h
73
PROOF. First of all, note that the extension E/F is also T ( E / F ) radical. Let p be an odd prime with Cp E T ( E / F ) . Then Cp E E , hence Cp t F by purity. If 1 + 6 E T ( E / F ) , then 1 + C4 E E , hence C4 E E, and so, C4 E F again by purity. Now apply Theorem 2.2.1 t o deduce that E/F is T(E/F)-Kneser. The same argument, or Corollary 2.1.10, shows that E/F is also G-Kneser. Then G = T ( E / F ) by Corollary 2.1.12.
R.ecal1 that a finite extension EIF is said t o be Cogalois if E/F is a radical extension such that ICog (E/F)I = [E : F].
Since @ = 7, we deduce that ord(f) is 1 or q. If we would have ord(C) = g, then the elements 7, . ,@-I of the group Cog(E/F) would be distinct. By Corollary 2.1.10, this would imply that the elements I , C, . . . , CQ-' of E would be linearly independent over F , which is a contradiction. Therefore, we must have ord(C) = 1, i.e., C E F * . If q = 4, we may again assume that Char ( F ) # 2. Since l + C - (1+C) = 0, the elements 1, C, 1 C of E are linearly dependent over F , hence, by Corollary 2.1.10, the elements 1, C, 1 C of Cog(E/F) cannot be distinct. But, any equality between two such cosets implies that ( E F, as desired.
c..
+
-+
THEOREM 3.1.7 (THE GREITHER-HARRISON CRITERION).The following assertions are equivalent for a finite extension E I F . (1) E/F is Cogalois. (2) E/F is radical, separable, and pure.
(2): Assume that EIF is a Cogalois extension. Then, PROOF. (1) by definition, it is radical. We are going to show that E/F is separable. Assume that E/F is not separable. Then necessarily Char(F) = p > 0 and [ E : F ] is divisible by p, because the degree of any finite nonseparable extension EIF is the product between its separable degree and a power of the characteristic of F (see 1.2.8). Since E/F is Cogalois, we have I Cog(E/ F ) I = [E : F ] , hence ICog(E/F)I is divisible by p. Then, by Lemma 1.4.11, the finite group Cog(E/F) contains an element E Cog(E/F) of order p, i.e., X $! F*and XP E F * . The field F is necessarily infinite, for otherwise, the extension E/F would be separable, which contradicts our assumption. Since X # F, we have p+X # F for all p E F, but (p+X)p = @+XP E F . n Observe t h 2 1 X # p2 A for every p1, p2 E F with p1 # p2. Indeed, p1 + X = p2 X implies that p l + X = a(p2+X) for some a E F * . If a # 1, then X = (1-11 - ap2) (a - I)-' E F , which is impossible. Hence a = I, and so, p1 X = p~ A, i.e., p l = p2. Thus, we showed that the finite group Cog(E/F) contains infinitely many elements of the form p + X with p E F, which is a contradiction. This proves that the extension E/F is separable.
Conversely, suppose that E/F is radical, separable and pure. Then Lemma 3.1.6 implies that EIF is T(E/F)-Kneser, i.e., Cogalois. 0 EXAMPI~ES 3.1.8. (1) The quadratic extension IF2 (X)/F2 ( X Z ) is radical, pure by Examples 3.1.4 ( I ) , but not separable, so it is not Cogalois. We have seen in Examples 2.1.5 (1) that the Cogalois group of this extension is a countably infinite group. (2) The quartic extension Q( a ) / Q is pure and separable, but it is neither radical nor Cogalois (see Proposition 3.2.6 (e)). (3) The quadratic extension Q(i)/Q is clearly a separable radical extension which is not pure, hence it is not Cogalois. We will show in Section 3.3 that Cog (Q(i)/Q) = { 1, i , 1 i, 1 - i ) = (1 2 ) E Z4. 0
A -
dl+
THEOREM 3.1.9 (THE GAY-V61mCRITERION). The following assertions are equivalent for a finite extension E I F . (1) E/F is Cogalois. (2) E / F is radical, separable, and satisfies the following conditions: for every odd prime p, CP E F whenever C, E T ( E / F ) , and C; E F whenever 1 C4 E T ( E I F ) . (3) EIF is radical, separable, and 6, $! E \ F for every p E P.
PROOF. (1) (3) follows immediately from the Greither-Harrison Criterion and Lemma 3.1.2. (1) =+ (2): If EIF is a Cogalois extension, then it is T(E/F)-Kneser. i IVow apply the Kneser Criterion. (2) ==+ (1): By Remarks 2.2.2 (1) and the Kneser Criterion, we deduce that the extension E/F is T(E/F)-Kneser, i.e., it is Cogalois. 0
Now we are going t o show that the extension EIF is pure. Let C E p,(E) with q E P. We have to show that C E F . Assume first that q is an odd prime. If q = p = Char(F), then Cp = 1 implies that = 1 E F . If q # p = Char(F), we can assume that C # 1 (since 1 E F ) , and then 1 C . . CQ-' = 0, i.e., 1, C, . . . , are linearly dependent over F .
<
+ + +
<Q-'
74
Chapter 3
3.2 Example and properties of Cogalois extensioris
75
REMARK 3.1.10. By the Greither-Harrison Criterion, any Cogalois extension is separable. However, a Kneser extensions is not necessarily separable, as the following example shows: F = IF2 ( X 2 ) , E = IF2 (X), G = F * ( X ) . The extension E / F is G-Kneser since IG/F*I = ord(X) = 2 = 0 [ E : F ] , but it is not separable.
A
(2) If E/F is a Cogalois extension, then E I K and K I F are both Cogalois extensions. (3) If E/F is a radical extension, and E I K , K / F are both Cogalois extensions, then E/F is a Cogalois extension. (4) If E/F is a Cogalois extension, then the groups Cog(E/K) and Cog ( E lF)/Cog ( K / F ) are canonically isomorphic.
PROOF. (1) Clearly, the canonical map

Cog ( E / F ) 4 Cog ( E / K ) , x F * I+ x K *
3.2. Examples and properties of Cogalois extensions
is a group morphism with kernel Cog ( K I F ) . This shows that the sequence
1
The aim of this section is t o present a series of examples, and to establish the basic properties of Cogalois extensions.
-+
Cog ( K I F )
-+
Cog ( E I F ) -+ Cog ( E / K )
of Abelian groups is exact. (2) Suppose that E/F is Cogalois. Then E/F is radical, separable and pure by Theorem 3.1.7, hence E / K is also radical, separable and pure. Again by Theorem 3.1.7, we conclude that E I K is a Cogalois extension. Now, we are going to show that K I F is also Cogalois. Since EIF is a Cogalois extension, the group Cog(E/F) is finite, hence its subgroup Cog(K/F) is also finite. Let r = ICog(K/F) I. If 6, . . . , are all the elements of Cog(K/F), then using again the fact that E/F is a Cogalois extension, we deduce by Corollary 2.1.10 that the subset {xl , . . . ,x, ) of K is linearly independent over F , hence By ( I ) , there exists a canonical monomorphism of groups %(EIF)lCog(K/F) hence
,
I
EXAMPLES 3.2.1. (1) Any finite G-radical extension E/F with E a subfield of R is pure by Examples 3.1.4 (I), hence it is Cogalois by the Greither-Harrison Criterion. Notice that for such an extension E/F we have Cog ( E / F ) = G / F * in view of Lemma 3.1.6. For example consider the extension Q("fi,... , Q / ) , al . ; " , where r E N*, n l , . . . ,n,, a l , . . . ,a, E N*, and " 6 is the positive real i r . This is a G-radical Cogalois ni-th root of ai for every i , 1 ... , hence its Cogalois group is extension, where G = Q precisely Q ( " f i , . . . , ".C/a,)/Q?.
< <
7 L * ) ,
( l L f i ,
(2) A quadratic extension Q ( J ~ ) / Q where d # 1 is a square-free integer is Cogalois if and only if d # -1, -3 (see Corollary 3.3.3). (3) By Exercise 1, for any odd prime p > 0 and any n E W ,the radical extension Q(Cpn )/Q(Cp) is pure, hence it is Cogalois by the GreitherHarrison Criterion. (4) By Exercise 4, Q(C9, fi)/Q)(C3) is a Galois and Cogalois extension 0 which is not Abelian. Next, we will investigate the property of an extension being Cogalois in a tower of fields.
Cog(E/K),
I C o d E / F >I /ICog(K/F) I ICog(E/K) I. Using the facts that EIF and E / K are Cogalois extensions, we obtain [ E : F]/ICog(K/F)I This implies that 1/ICog(K/F)I 6 1 / [ K : Fl,
<
< [ E :K ] = [ E : F ] / [ K : F ] .
PROPOSITION 3.2.2. The following assertions hold for a tower of fields FGKCE. (I) There exists a canonical exact sequence of Abelian groups
i.e., ICog(K/F)I 2 [ K : F l . Since we have shown that the opposite inequality also holds, we deduce that ICog(K/F)I = [ K : F ] .
76
Chapter 3
3.2 Example and properties of Cogalois exterisioris
77
This shows that ... is a vector space basis of K over F . In particular K = F (xl ,. . . ,x,), and since {xl , . . . , x,) C T ( K / F ) , it follows that K / F is a radical extension. Consequently, K / F is a Cogalois extension by the Greither-Harrison Criterion. We present below an alternative proof of the fact that K / F is a Cogalois extension. A piece of this proof will be used in the proof of Theorem 3.2.3 (1). For simplicity, set G = T ( E / F ) . Since E I K is Cogalois, it is T(E/K)-Kneser. But G K * = T ( E / F ) K * T ( E / K ) , and E = F ( G ) = K ( G K * ) , hence the extension E / K is GK*-radical, separable and pure. Consequently, G K * = T ( E / K ) by Lemma 3.1.6, i.e., E / K is GK*-Kneser, hence [E : K ] = I(GK*)/K*I = IG/(K* nG)I. Thus
{c, ,c)
THEOREM 3.2.3. The following statements hold for a finite Cogalois extension E/ F. (1) The maps - n T ( E / F ) : & + C and F ( - ) : C ---+ & are isomorphisms of lattices, inverse to one another, where E = { K I F C K , K subfieldof E ) and C = { H I F * H < T ( E / F ) ) . (2) For every intermediate field K E & one has K = F ( T ( K I F ) ) . (3) For every subgroup H E C one has C o g ( F ( H ) / F ) = H I F * .
<
<
PROOF. (1) For every H E C and K E & we have by Proposition 2.1.11, and by the alternative proof of point (2) in Proposition 3.2.2. These two relations mean precisely that the order-preserving maps - n T ( E / F ) : & -+ C and F ( - ) : C --+ & are bijections, hence isomorphisms of lattices, inverse t o one another. (2) and (3) are consequences of (1) and of the following simple facts: for every H E C and every K E & one has K
On the other hand, F ( K * n G ) / F is K * n G-Kneser by Corollary 2.1.10, so [ F ( K * n G ) : F] = I(K*nG)/F*I. Since F ( K * n G ) 5 K we deduce that K = F ( K * n G). Consequently, K / F is K * n G-Kneser. But K * n G = K * n T ( E / F ) = T ( K / F ) . Thus K I F is T(K/F)-Kneser, i.e., K / F is a Cogalois extension. (3) Suppose that E / K and K I F are both Cogalois extensions. Since E / K and K / F are separable extensions, so is E I F . Since E I K and K / F are pure, so is also EIF by Proposition 3.1.5. But E/F is a radical extension by hypothesis, hence E/F is Cogalois by Theorem 3.1.7. (4) By ( I ) , there exists a canonical monomorphism of Abelian groups. Observe that since all extensions E / F , K I F , E / K are Cogalois, one has
n T(E/F) = T(K/F)
and H = F ( H ) n T ( E / F ) = T ( F ( H ) / F ) .
P R ~ P O S I T3.2.4. I O N Let E1/F and E2/F be Cogalois extensions which are subextensions of a n extension E/F. Then, the extension El E2/F is Cogalois if and only if it is pure. I n this case, the following statements are equivalent. (1) T ( E I I F ) n T ( E 2 / F ) = F * . (2) The canonical map
defined by a ( x l F * , x 2 F * ) = x1n2F*,X I E E l , E E 2 / F , is a monomorphism of groups. (3) The canonical map a in (2) is a n isomorphism of groups. (4) The fields El and E 2 are linearly disjoint over F . (5) El n E2 = F.
PROOF. We have El = F ( T ( E 1 / F ) ) and E2 = F ( T ( E 2 / F ) ) , hence El Ez = F ( T ( E l I F ) U T ( E 2 / F ) ) . Thus El E2/ F is a radical extension. Since El /F and E2/ F are separable, so is also El E2/ F . Therefore, by the
= [ E :F ] / [ K : F] = [ E : K ] = ICog(E/K)I,
hence cp is also surjective. Thus
is an isomorphism of groups.
An easy consequence of Proposition 3.2.2 is the following fundamental result, which essentially says that any Cogalois extension is an extension with Cogalois correspondence. A much shorter proof, involving the nPurity Criterion will be given in Section 7.5.
78
Chapter 3
3.2
Example and properties of Cogalois extensions
79
Greither-Harrison Criterion, the extension El E2/Fis Cogalois if and only if it is pure. (1) ==+ (2): We are going to show that the group morphism a is injective. As always throughout this monograph, for any x E E we denote by P the coset x F * in the quotient group E*I F * . If (It, 9) E Ker(a), then G j = 1, hence xy E F*.Since x E T (El I F ) it follows that y E T ( E l I F ) . But y E T ( E 2 / F ) , so y E T ( E I / F ) n T ( E 2 / F ) = F * . This implies that x E F * . Consequently, (It, 9) = (7,i),hence Ker(a) = {(T, 7)). This shows that a is injective.
A
REMARKS 3.2.5. (1) The compositum of two Cogalois subextensions of an extension is not necessarily Cogalois, by Exercise 5. (2) By Exercise 6, the conditions (1)-(5) in Proposition 3.2.4 are also equivalent t o the following one:
The groups Cog(E1E 2 / F ) and Cog(E11F) x Cog(E2lF) are isomorphic, but n o t necessarily via the canonical isomorphism a . 0
We end this section by showing that the property of an extension being radical, Kneser, or Cogalois is, in general, not transitive. This will follow immediately from the next proposition investigating the quartic extension Q(rn)/Q.
(2) ==+ (3): We have t o prove that. a is surjective. To do that, observe that since El E2IF is Cogalois, one has
PROPOSITION 3.2.6. The following assertions hold.

(a) ~ ( 4 5 is ) a subfield of the field Q ( d m ) . But a is a monomorphism, hence
: Q ( d ? ) ] = 2 and : Q] = 4. (b) (c) Q ( d m ) / Q is not a Galois extension. (d) Q ( d m ) / Q ( d T ) and Q(,/Z)/Q are both Cogalois extensions. (e) Q(d1$fi)/Q is neither a radical extension, nor a Kneser exten-
[()(dm)
[ ~ ( d m )
Therefore a is surjective, hence an isomorphism. (4): By the definition of linearly disjoint extensions (see l.2.3), (3) we have to show that there exists a vector space basis Bi of Ei over F , i = 1, 2, such that BBI B2 = { xlx2 1 x1 E B1, x2 E B2 ) is a vector space basis of El E2 over F. Let Bi be a set of representatives in T ( E i / F ) of the factor group Cog(Ei/F) = T ( E i / F ) / F * , i = 1, 2. Then Bi is a vector space basis of Ei over F , i = 1, 2, by Corollary 2.1.10. Since Cog(Ei/F) = { Fi 1 xi E Bi ), i = 1, 2, and a is an isomorphism we deduce that and for every X I ,yl E B1, x2, y2 E B2. In other words, B I B 2 is a set of representatives of Cog(E1E2/F). By Proposition 2.1.8, it follows that B1B2 is a basis of the vector space ElE2 over F , as desired. (4) ==+ (5) and (5) ==+(1) are obvious.
sion, nor a Cogalois extension.
order
PROOF. For simplicity, set
(a) We have and E = K(8).
fi = O2 - 1 E Q(8) = E , hence
K = Q(JZ)
E Q(8)
= E,
- 2X2- 1 E Q[X], (b) Since 8 is a root of the irreducible polynomial x4 we deduce that [ E : F ] = 4, and this clearly implies that [ E : K ] = 2.
(c) The conjugates of 8 over Q are the roots of its minimal polynomial
X4 - 2X2 - 1 over Q Two of them are the nonreal complex roots of the equation x2 = 1 - \/2, SO they do not belong to E 5 R This shows that the extension EIF is not normal, so it is not Galois.
80
Chapter 3
3.2 Bxarnple and properties of Cogalois extensions
81
(d) Since E is a subfield of the field R of real numbers, it follows that the extensions K / F and E / K are both pure by Examples 3.1.4 (1). On the other hand, the extensions K I F and E / K are both separable and radical. So, in view of the Greither-Harison Criterion, K / F and E / K are both Cogalois extensions. (e) The extension EIF is pure, again by Example 3.1.4 (1). Since E/F is clearly a separable extension, then E/F is a Cogalois extension if and only if it is radical by the Greither-Harison Criterion. Since any Kneser extension is necessarily radical, it is sufficient to prove only that E/F is not a Cogalois extension. Assume that E/F is a Cogalois extension, and look for a contradiction. Since [ E : F] = 4, then Cog(E/F) is a group of order 4 by the definition of the concept of Cogalois extension, hence this group is isomorphic either to Z 2 X Z2 t o Z4.
Case 1: If Cog(E/F) Z Z 2 x Z2, then there exist ,L?, 7 E
an intermediate field, this implies that necessarily Q(JE) = Q(J2), hence \/5 = k f i for some k E Q , and then 2 = a k 2 . Since E = Q(0) = Q ( y E ) , there exist a , b, c, d E Q such that
We can also write (1) as (2) 0- (a+c&)= *(b+d&).
Squaring (2), we obtain
O2 --2O(a+c&)
+ ( a + ~ f i )= ~ &(b+d&)2.
Since O2 E Q((J?) = Q(JZ ), we deduce that we must have a c& = 0, for otherwise, it would follow that 0 E Q(JZ) = Q(J?), which is a contradiction. Thus, (2) becomes
(&
such
that Then, by Corollary 2.1.10, we deduce that { 1, fi, fi, space basis of E over Q. In particular, it follows that
(3)
0 = &(b+d&),
which can be also written as
) is a vector
But Q(J4, &)/Q is a Galois extension, hence so is Q ( J m ) / Q , which contradicts (c). Thus, this case cannot occur.
Case 2: If Cog(E/F) Z Z4, then there exists a E U &
since f i = k. From (4) we deduce that
a/
such that
Again by Corollary 2.1.10, { 1, fi, fi2, E over Q. In particular, it follows that
@%3
} is a vector space basis of
(bk d ~ 2 ) ~ E Q:= (1 + J Z ) 2 Then, u coincides with its conjugate in the quadratic extension Q(JZ)/Q, i.e., (bk ~ J (l+JZ)'
Since [ E : F] = 4, we have fi $ Q. Clearly, Q ( J 5 ) is a proper subfield of Q(g/&) = E . Since the group Cog(E/F) has a unique proper subgroup, we deduce by Theorern 3.2.3 (1) that E/F has a unique proper intermediate field. Since Q(J2) is also such
Z) (bk ~ -d
~ i ) ~
.
82
Chapter 3
3 . 3 The Cogalois group of a quadratic extension
83
This implies that (bk hence (bk + dJZ) (JZ - 1) = (bk - dJZ) ( J Z After easy calculations we obtain b2k2 - 4bdk which can be also written as
(5)
+d d )
(bk - d d ) '
1+JZ
JZ-1
'
follows. It is easy to show that fi @ Q(Jj7 ), hence [ Q(JB, J;;; ) : Q ] = 4. and then E/F is a radical extension, This implies that E = Q(,@, which contradicts (e). If Cog(E/F) 2 Z4, then there exists cu E such that
a)?
+ 1).
+ 2d2 = 0,
6 Q( JZ ), it follows that { 1,*, 2 , Since } is a vector space basis of E over F . Thus, EIF is a radical extension, which again contradicts (e). This completes the proof of (f).
= @ (g) Assume that o r d ( 7 3 ) is finite, say n. Then, for some a E so it would follow that the extension Q ( J ~ ) Q / is a subextension of the Cogalois extension Q( *)I$ hence it would be 0 itself Cogalois, which contradicts (e).
(bk - 2d)2 = 2d2.
E Q, which is We claim that d = 0, for otherwise, (5) would imply that a contradiction. Then, (5) yields bk = 0, hence b = 0. Thus, (3) becomes 8 = 0, which is a contradiction. This completes the proof of the fact that E / F is not a Cogalois extension.
a,
Ja
(f) The inclusion
COROI~LARY 3.2.7. The property of an extension being radical, Kneser, 0 or Cogalois is, in general, not transitive.
is clear. To prove the equality it is sufficient to show that /Cog ( E / F ) ( = 2. To do that, observe that, by Proposition 3.2.2 (I), there exists a canonical monomorphism of groups By (d), the quadratic extensions E I K and K I F are both Cogalois. Then hence necessarily ICog ( E I F )I E {2,4). Since EIF is not a Cogalois extension by (e), we are going to show that we cannot have ICog ( E I F )I = 4. This will imply that /Cog (E/F)I = 2, and we will be done. Assume that [Cog(E/F)I = 4. Then, as in the proof of (e), two cases arise: Cog(E/F) r Z2 x Z2, or Cog(E/F) r Z4. In the first case, there exist 8, -y E U& such that
= { 7, }. C % ( E I F ) = (&, We claim that { 1, &, ) is a vector space basis of E over F. Since EIF is not a radical extension, we cannot apply Corollary 2.1.10 t o deduce it, as we did in the proof of (e). However we can prove the clai~n as
3.3. The Cogalois group of a quadratic extension

For any G-radical extension E I F , the group G I F * is a subgroup of the Cogalois group Cog(E/F) = T ( E / F ) / F * of E I F . Hence, the effective description of the Cogalois group of a given extension is an important problem in the investigation of radical extensions, which is also of independent interest. In general, the concrete calculation of the Cogalois group of a given extension is quite hard. The aim of this section is to provide a very explicit description of the Cogalois group of any quadratic extension of Q. 6 As usually, we shall denote the imaginary unit d by i& for any d E Q$.
fl E
(C
by i, and
a,
,m/v
z, 3, fi
LEMMA 3.3.1. Let d # 1 be a square-free rational integer, and let W denote the group of roots of unity p ( ~J d ( ) in ~ ( d d ) Then .
(1) W = p2(@) = { H ) if d # -1, -3. (2) W = p4((C) = { f1, i i ) if d = -1. (3) w = p6(@)= {&I, i ( 1 i 4 ) / 2 , ~ ( 1 i &)/2)
if d = -3.
84
Chapter 3
3.3 'I'he Cogalois group of a quadratic exter~sior~
85
PROOF.If the quadratic field Q(J d ) is real, that is, if d > 0, then W consists only of real roots of unity, hence W = {-1,l). If the quadratic field Q( 4 2 ) is imaginary, that is, if d < 0, then W exhausts all the units of the ring of integers A of the given quadratic field, and they are precisely those elements of A having norm 1. An easy computation of the norm of an element in A establishes the result. More details can be found, e.g., in 17 Ribenboim [89, p.1311.
i) If [ = k 1 , then ( t ) implies that a2 E Q, i.e., a 2 - b2 ab = 0. Then a = 0, hence a = bi, and thus 6 = i.

A
+ 2abi E Q, so
ii) If ( = f i, then ( t ) implies that i a 2 E Q, i.e., -2ab f (a" b2)i E Q, hence a 2 - b2 = 0, and so a = f b. If a = b, then we have a = b(l + i), hence 2 = 1 i, and if a = -b, then we have cu = -b(l - i), hence = 1 - i. Thus
n n -
PROPOSITION 3.3.2. Let E = ~ ( d d ) where , d # 1 is a square-free integer. Then

(1) Cog (EIQ) = (& ) Zz if d # -1, -3. (2) Cog(E/Q) = ( l + i ) ? Z q if d = - 1 . (3) c o g (E/Q = ( i a . (1 + i a ) ) Zf3 if d = -3.
A
4
L
Cog(Q(i)/Q) = { I , i, l + i , 1 - i } = ( l + i ) 2 Z 4 .
- + - -
-h
Case 3: d = -3. We have a = a b i f i and a2 = a2 - 3b2 By Lemma 3.3.1, ( E { f1, f ( l i&)/2, f ( l - i f i ) / 2 }.
+ 2abifi.
E'
i) If ( = f1, then ( t ) irnplies that a2 E Q, i.e., a2 - 3b2 2 a b i f i E Q, hence ab = 0. Then a = 0 and a = b i f i , hence
PROOF. Let a E T(E/Q), a = a + b& with a , b E Q, and consider 2 in Cog (Q( Jd)/Q) = T(E/Q)/Q. Clearly, a E Q if and only its coset i if 6 = 1. Kow suppose that a # Q, i.e., b # 0. Since a E T ( E / Q ) , there exists n E N,n > 1 such that an = c E Q, hence a is a root of the polynomial f = X n - c E Q[X]. The roots of this polynomial have the form a(, where ( E p,,(C). The minimal polynomial Min(a,Q) of a over Q has degree 2 since E = Q(a), and is a divisor of f in Q[X]. Consequently, the roots of Min(a,Q) are a and (a, where ( E p(C). Moreover, (0 E E since the extension E / Q is normal. But C = ((a)/a, hence ( E E n p(C) = p ( E ) = W. On the other hand, the product of the roots of Min(a, Q) is a rational number, hence
2 = ifi.
/':
. .
(1
+ i&)(a2
ii) If ( = k(1 + i&)/2, then we have ( I + ifi)a2 E Q by (t), i.e., - 3b2 2abifi) E Q. Thus
a" - 3b"
6ab
+ (a"
+
3b2
+ 2 a b ) i h E Q,
and so, a " 3b" 2ab = ( a - b)(a 3b) = 0. It follows that a = b or a = -3b. If a = b, then a = b(1 + i d ) , hence 6 = 1+ ifi, and if a = -3b,
$1
then a = b(-3
+ ifi),
hence 6 = -3
+
+
i f i = i d . (1 + i f i ) .
..f
'F
4:.
iii) If [ = f(1 - i&)/2, then we have (1 - i & ) a 2 E Q by ( t ) , i.e., (1 - i&)(a2 - 3b2 2 a b i A ) E Q Thus
Row, we are going to investigate all the cases.considered in the staternent of the proposition.
,y
3b) = 0. It follows that a = - b or and so, -a2 3b2 2ab = (a b) (-a a = 3b. If a = -b, then a = -b(l - i & ) , hence 2 = 1 - i h , and if a = 3b, then a = b(3 i& ), hence 6 = 3
+ +
+
Case 1: d # -1, -3. We have ( E (-1, 1 ) by Lemma 3.3.1, and (t) irnplies that a" Q, i.e., a' + db2 + 2ab E Q. Then ab = 0. Since b # 0 , we necessarily have a = 0 , hence a = b&, and then, 6 = &. It
A . .
F: {
i!.
+i
h = i f i . (1 - i f i ). Consequently,
follows that
c ~ ~ ( E / Q= ){ F , J ; i ) = ( J ; i ) 2 ~ 2 .
Case 2: d = -1. We have a = a + bi and a2 = a 2 - b2 Lemma 3.3.1, C E (1, -1, i, - i ) .
+ 2abi. By
COROLLARY 3.3.3. The following statements are equivalent for a square-free rational integer d # 1.
86
Chapter 3
87
(1) Q(JTi)/Q is a pure extension. (2) Q(&)/Q is a Cogalois extension. (3) d # -1, -3.
PROOF. By definition, Q(&)/Q is a Cogalois extension if and only if /Cog (Q(dd)/Q)l = 2, i.e., by Proposition 3.3.2, if and only if d # -1, -3, (3). The equivalence (1) w (2) follows from the Greitherso (1) Harrison Criterion. 0
6. Let El /F and E2/F be subextensions of an extension E / F , and assume that the extension El E2/F is Cogalois. Prove that the groups Cog(E1/ F ) x Cog(E2I F ) and Cog(EIE2/ F) are isomorphic if and only if the fields El and E2 are linearly disjoint over F.
7. Let E/F be a finite extension with ICog(E/F)i = [ E : Fj = n. (a) Show that E / F is a Cogalois extension whenever n is a prime number, or n = 4 and E is a subfield of R. (b) Is EIF a Cogalois extension for arbitrary n?
REMARK 3.3.4. Any extension E/F of algebraic number fields has a finite Cogalois group. This will be proved in Section 6.1 using the description of the Cc*dois group of a finite Galois extension by means of crossed homomorphisms (see Corollary 6.1.4). 0
8. Show that the extension lF4/F2 is not Cogalois.

9. Let n E N*. Prove that the extension Q(c,,)/Q is Cogalois if and only if n = 1 or n = 2.
10. Show that any quadratic extension E I F , with E a subfield of Cogalois. 3.4. Exercises to Chapter 3 1. (Greither and Harrison [63]). Prove that for any odd prime p and any n E N* , the extension Q(cpn)/Q(sp) is pure. (Hint: Show that the fields Q(sprL ) and Q(cq) are linearly disjoint over Q whenever q # p is a prime or q = 4.) 2. Show that the result in Exercise 1 fails for p = 2.
3. ( Greither and Harrison [63]). Let F = Q(C3) and E = F (Cg,fi). Prove that the extension EIF is pure.
B is
11. Prove that the extension Q ( J m ) / Q Kneser, nor Cogalois. 12. Prove that ~ o g ( Q ( d r f i ) / Q ) = 13. Let v,. =
is neither radical, nor
{T,3).
Fr = Q(u,),
i
J+
1 a n , / d 1+
7
2.
4. (Greither and Harrison [63]). Prove the following statements for the fields F = Q ( 6 ) and E = F(Cg, fi). (a) E/F is a Galois extension of degree 27, with Gal(E/F) = ( a , ~ ) and r o a = a 4 o r , where a ( S ) = b f i , a(Ss) = ~ s , r ( $ 5 )= $5, (b) E/F is a Cogalois extension with
= Q.
radicals Show that F,/Q is a non Galois extension of degree 2' which is neither radical, nor Kneser, nor Cogalois.
r
14. With the notation of Exercise 13, determine Cog (F,/Q)
15. With the notation of Exercise 13, check whether or not, for a given is Cogalois. r E N* , the quartic extension FT+l
16. Let d E Q
\ @. Show that
if - d, -3d $ @ if - d ~ @ if - 3 d ~ @ .
cog ( ( d ) ) = 5. Let F = 0 , El = Q ( J Z ) , and E2 = Q(JZ). Show that E1/F and E2/F are Cogalois extensions, but their compositum El E2/F is not pure, so, not Cogalois.
88
Chapter 3

Section 3.1. To the best of our knowledge, the term of "Cogalois extension" appeared for the first time in the literature in 1986 in the fundamental paper of Greither and Harrison [63], where Cogalois extensions were introduced and investigated. A finite extension E/F is called conormal (resp. coseparable) by Greither and Harrison if ICog(E/F)J [ E : Fj (resp. if EIF is radical), and it is called Cogalois if it is both conormal and coseparable. Their pioneering work on Cogalois extensions was continued in 1991 by Barrera-Mora, Rzedowski-Calderh, and Villa-Salvador [30]. The Cogalois extensions, as well as the Galois extensions are related t o the Galois H-objects from the theory of Hopf algebras in a dual manner (see Section 10.2 for connections between Cogalois Theory and Hopf algebras). Theorem 3.1.7 is due t o Greither and Harrison [63]. The easy implication ( I ) (2) follows the original proof in [63], while the difficult one is a simplified proof of Albu and Nicolae from [19], based on Lemma 3.1.6, which, in turn, is an immediate consequence of the Kneser Criterion and Proposition 2.1.11. Note that the original proof in Greither and Harrison [63] is much longer and uses some cohomology of groups. Theorem 3.1.9 is an extended reformulation of Theorem 1.7 in Gay and Vdez [61]. Kote that Gay and Vdez proved only the implication (3) ===+ ( I ) , but they omitted in the statement of their theorern the indispensable separability condition of the given field extension. This section basically follows the outline in Albu and Nicolae [19], although there are several extensions, simplifications, and improvements.
<
CHAPTER 4
STRONGLY KNESER EXTENSIONS

This chapter contains the main results of Part I of the monograph. The notions of Cogalois connection, strongly G-Kneser extension, and GCogalois extension are introduced. The last ones are those separable GKneser extensions E/F for which there exists a canonical lattice isomorphism between the lattice of all subextensions of E / F and the lattice of all subgroups of the group G / F * . A very useful characterization of G-Cogalois extensions in terms of n-purity is given, where n is the exponent of the finite group G I F * . Using this characterization, we will show in Chapter 7 that the class of G-Cogalois extensions is large enough, including important classes of finite extensions. We show that a separable G-Kneser extension E/F is G-Cogalois if and only if the group G I F * has a prescribed structure. As a consequence, the uniqueness of the group G is deduced. This means that if the extension E/ F is simultaneously G-Cogalois and H-Cog alois, then necessarily G = H. Consequently, it makes sense t o define the Kneser group of a G-Cogalois extension E/F as the group G I F * , which will be denoted by Kne(E/F). Finally, we investigate almost G-Cogalois extensions, introduced by "am-Estrada, Barrera-Mora, and Villa-Salvador [79j under the name of pseudo G-Cogalozs extensions. These are the finite separable G-radical extensions E/F with G/F*-Cogalois correspondence and G I F * finite. Thus, the G-Cogalois extensions are precisely those almost G-Cogalois extensions which are also G-Kneser. Throughout this chapter E/F will denote a fixed extension and G a group such that F* 6 G E*. We shall also use the following notation:
Section 3.2. Proposition 3.2.2 and Theorem 3.2.3 are due t o Greither and Harrison [63j. Proposition 3.2.4, which is an extended version of Proposition 10 in Barrera-Mora, Rzedowski-Calderbn, and Villa-Salvador [30] is the finite case of Proposition 3.6 in Aibu [a]. Proposition 3.2.6 is taken from Albu [7]. Section 3.3. The determination of the Cogalois group of a finite field extension is quite hard. In the literature there are several rather theoretical, but less concrete approaches t o this matter, especially due t o Acosta de Orozco and Vklex [2] and Gay and Vklw [GI]. This section essentially follows the presentation in Albu, Kicolae, and Term [23] and Term [104].
<
E := Intermediate ( E / F ) = { K I F C_ K , K subfield of E ).
Chapter 4
4.1 Galois and Cogalois cor~nections
4.1. Galois and Cogalois connections
is a Galois Connection. Observe that if
In this section we present the dual concepts of Galois connection and Cogalois connection for arbitrary posets, and related t o them, the concepts of closed element. Then, we illustrate them with examples occurring in Field Theory, and introduce the concepts of field extension with Galois correspondence and field extension with Cogalois correspondence. DEFINITION 4.1.1. A Galois connection between the posets (X, ,<) and (Y, 6 ) is a pair of order-reversing maps a :X
is a Galois connection, then Y - X

__C
---+
Y and p : Y
-+ X
is also a Galois connection. However, this sort of symmetry does not hold for a Cogalois connection. Let
satisfying the following conditions:
Whenever we have a Galois connection as in Definition 4.1.1, we shall use the notation
CY
__C
Y.
be a Galois or Cogalois connection. If x E X (resp. y E Y), then we shall briefly denote by x' (resp. y') the element a ( x ) (resp. P(y)) ; also, we shall use the notation:
:= (x')',
D
If the maps a and ,f3 are both order-preserving instead of orderreversing, we obtain a Cogalois connection between X and Y. hilore precisely, we introduce the following definition. D E F ~ N ~ T4.1.2. ~ O N A Cogalois connection between the posets (X, <) and (Y, 6 ) is a pair of order-preserving maps
a :X
x"' := (x")', y" .- (y')', y"'
:= (y")'.
An element z of X or Y is said to be a closed element of X or Y, if z = z". A closed element is also called Galois object (resp. Cogalois object) in the case of a Galois (resp. Cogalois) connection. We shall denote by (resp. Y) the set of all closed elements of X (resp. Y). The next result collects the basic facts on Galois and Cogalois connections which will be freely used throughout this monograph.
-+
Y and
p : Y --+ X
satisfying the following conditions:
If we denote by XOPthe opposite poset of X, then it is clear that

CY
x - Y
___t
P
is a Cogalois connection if and only if
PROPOSITION 4.1.3. W i t h notation above, the following assertions hold for a Galois or Cogalois connection between the posets X and Y. ( I ) z' = z"' for every element z of X or Y. (2) X = P ( Y ) and P = a ( X ) . (3) The restrictions G : +Y and : Y -+ X of a and 0 t o the sets of closed elements of X and Y are bijections inverse t o one another.
PROOF. We will examine only the case of a Cogalois connection. Let x E X and y E Y. Then, the conditions from Definition 4.1.2 can be shortly expressed as x" 6 x and y 6 y". Since the priming operation is
92
Chapter I
4.1 Galois and Cogalois connections
93
an order-preserving map, we have
< 5' and y' < (y")' = y"'. If we replace in the relation y < y" the element y E Y by xi, we deduce that x i 6 (xi)" = xi" , which proves that xi = xu'. Similarly, if we replace in the relation x" < x the element x E X by y' we deduce that (y')" = y"' < y' which proves that y' = y"'. This
xi'' = (xi')'
!
(6) The maps a and /? establish anti-isomorphisms of lattices, inverse t o one another, between the lattices Intermediate ( E I F ) and Subgroups (I') .
PROOF. The irnplications (1) =j (2) ==+ (3) ==+ (1)==+ (6) are well known (see 1.2.9).
The implications (6) ==+ (5) and (6) ==+ (4) are trivial. (5) a (3): If /? is surjective, then F = /?(A) = Fix(A) for some subgroup A of I?, i.e., F is a closed element of Intermediate ( E I F ) . (4) + (3): We know that F' = F"' = (F")', i.e., a ( F ) = a ( F U ) . Since a is injective, we deduce that F = F", hence F is a closed element 0 of Intermediate ( E l F ) . The prototype of a Cogalois connection is that canonically associated with any radical extension. Let EIF be an arbitrary G-radical extension. Then, the maps
establishes (1). Now, let x E X. Then x = x" = (xi)' = /?(xi) E /?(Y). Conversely, if x E /?(Y), then x = b(y) = y' for some y E Y. But y' = y"' = (y')" by ( I ) , so y' = x is a closed element of X. This shows that fT = /?(Y). In a similar manner one proves the other equality in (2). Then y = a ( x ) = xi for some x E X by (2), To prove (3), let y E 7. hence (a0 p)(Y)= a(/?(a(x)))= x'" = xi = y. 0 Similarly, o z)(x) = x for all x E
(p
x.
The most relevant example of a Galois connection, which actually originated the name of this concept, is that appearing in Galois Theory. Let E/F be an arbitrary field extension, and denote by I? the Galois group G a l ( E / F ) of E I F . Then, it is easily seen that the maps
a : Intermediate ( E I F ) + Subgroups (I?), a ( K ) = Gal(E/K),
x : Intermediate ( E I F ) -+
and
Subgroups ( G I F * ) , x ( K ) = ( K n G ) / F * ,
w : Subgroups ( G I F * ) + Intermediate ( E I F ) , w(H/F*) = F ( H ) ,
/? : Subgroups (I?) + Intermediate ( E I F ) , b ( A ) = Fix(A),

yield a canonical Galois connection between the lattice Intermediate ( E I F ) of all intermediate fields of the extension EIF and the lattice Subgroups (I') of all subgroups of I?. We will call it the standard Galois connection associated with the extension E I F .
establish a Cogalois connection between the lattices Intermediate ( E I F ) and Subgroups (GIF*). We will call it the standard Cogalois connection associated with the extension E I F . Notice that, in contrast with the standard Galois connection which is associated with any extension, the standard Cogalois connection can be associated only with radical extensions. Clearly, the lattice Subgroups ( G I F * ) is canonically isomorphic t o the lattice G={HIF*<H<G), so, if we denote by & the lattice of all intermediate fields of the extension E I F , then the Cogalois connection described above is essentially the same with the following one, which is also called the standard Cogalois connection associated with E I F :
PROPOSITION 4.1.4. W i t h the notation above, the following assertions are equivalent for a finite extension EIF with Galois group I?. (1) EIF is a Galois extension. (2) Every intermediate field of the extension EIF is a closed element i n the standard Galois connection associated with E I F . (3) F is a closed element i n the standard Galois connection associated with E I F . (4) The map a is injective. (5) The map P is surjective.
E
where
1CI
Y '
___t
Gl
Chapter 4
4 . 2 Strongly G-Kneser extensions
95
The considerations above naturally lead us to define the following dual concepts.
' is said to DEFINITIONS 4.1.5. An extension EIF with Galois group I be an extension with I'-Galois correspondence if the standard Galois connection associated with E/F yields a lattice anti-isomorphism between the lattices Intermediate ( E I F ) and Subgroups (I') . Dually, a G-radical extension E/F is said to be an extension with G/F*-Cogalois correspondence if the standard Cogalois connection associated with E/F gives rise to a lattice isomorphism between the lattices Intermediate ( E / F ) and Subgroups (GIF*). 0
R.EMARK 4.1.6. Note that any finite extension E/F with I'-Galois correspondence is necessarily a Galois extension by Proposition 4.1.4. Consequently, the equality [ E : F ] = IGal(E/F)I is a consequence of the fact that E/F is an extension with I?-Galois correspondence. Conversely, if a finite extension E/F is such that [ E : F ] = IGal(E/F)I, then E/F is necessarily a Galois extension (see Exercise 21, Chapter 1). This is not the case for finite extensions E/F with G/F*-Cogalois correspondence; namely, for such extensions, the equality [ E : F ] = IG/F*I, saying precisely that EIF is G-Kneser is, in general, not a consequence of the fact that EIF is an extension with G/F*-Cogalois correspondence. We will examine this situation more closely in Section 4.5. 0
PROPOSITION 4.2.1. Let E/F be a finite G-Kneser extension, and let K be an intermediate field of E I F . Then, the following assertions are equivalent. (1) K / F is H-Kneser for some H E G. (2) K I F is K * n G-Kneser. (3) E / K is K*G-Kneser.
PROOF. (1) ==+ (3): Suppose that K / F is H-Kneser for some H E G. Then K = F ( H ) , [ K : F] = IH/F*I, and K * n G = K n G = H by Proposition 2.1.11. Clearly E = F ( G ) = K ( G ) = K ( K * G ) . But
(K*G)/K* FZ G / ( K * n G) = G I H , and (K*G)/K* is finite since G / F * is finite and F* H G. Thus, K * K * G T ( E / K ) , hence E / K is a radical K*G-extension. Moreover,
<
<
<
<
[ E : K ] = [ E : F ] / [ K : F ] = IG/F*I/IH/F*I
= IG/HI = I(K*G)/K*I,
hence E / K is K*G-Kneser.
(3)
hence
=j
(2): Suppose that E / K is K*G-Kneser. Then
4.2. S t r o n g l y G-Kneser extensions
As we noticed in Remarks 2.1.14 (I),a subextension of a Kneser extension is not necessarily a Kneser extension. In this section we introduce and investigate strongly G-Kneser extensions; namely, those G-Kneser extensions E/ F such that every subextension K / F of E/F is K * nG -Kneser. It turns out that such extensions are precisely the G-Kneser extensions with G IF*-Cogalois correspondence. Recall that throughout this chapter E/F will denote a fixed G-radical extension, and G:={HIFC<H<G), & := Intermediate ( E / F ) = { K I F C K , K subfield of E ).
[ K : F] = [ E :F ] / [ E : K ] = IG/F*J/IG/(K*nG)I = [ ( K *n G ) / F * I . On the other hand, F ( K * n G ) / F is K * n G-Kneser by Proposition 2.1.11, so i F ( K * n G) : F] = [ ( K *n G)/F*I = [ K : F ] . Since F ( K * n G) K , we deduce that K = F ( K * n G). Observe that F* < K * n G < T ( K / F ) since F* 6 G 6 T ( E / F ) and T ( E / F ) n K * = T ( K / F ) , hence K / F is a K * n G-radical extension. Consequently, K / F is K * n G-Kneser, again by Proposition 2.1.11. C] (2) =j (1) is obvious.
PROPOSITION 4.2.2. Let F E K 2 E be a tower of fields, and let G be a group such that F* G E*. If K / F is K * n G-Kneser and E/ K is K*G-Kneser, then E/F is G-Kneser.
< <
PROOF. In order t o prove that the extension EIF is G-Kneser, we have t o check first that E/F is G-radical. To do that, observe that, by hypotheses, K = F ( K * n G) C F ( G ) , hence E = K ( K * G ) F ( G ) , and so, E = F ( G ) . Now, let g E G. Since the extension E I K is K*G-radical, ( K * G ) / K * is a torsion group, hence g " E K * for some m E W . Further,
96
Chapter 4
4.2 Strongly G-Kneser exterisioris
97
K I F is a K * n G-radical extension, so gmn = (g7n)n E F* for some n E K . This shows that G I F * is a torsion group, in other words, E J F is a G-radical extension. On the other hand, we have
= IG/(K* n G)I . I(K* n G)/F*I = (GIF'I, Thus, EIF is a G-Kneser extension.
which is a contradiction. H Thus, by Proposition 4.2.1, for every H with Q Q(JB)/Q is not a H-Kneser extension. However, Q( & ) / Q
< <
(m),
is clearly
0
Theorem 4.2.3 and Example 4.2.4 naturally lead t o the following definition.
Propositions 4.2.1 and 4.2.2 can be reformulated together as follows.
THEOREM 4.2.3. Let F C K C E be a tower of fields, and let G be a group such that F* G E*. Consider the following assertions. ( I ) K / F is K * G-Kneser. (2) E I K is K*G-Kneser. (3) EIF is G-Kneser. Then, any two of the assertions (1)-(3) imply the remaining one. 0 EXAMPLE 4.2.4. Denote by one of the complex roots, say &(I i)/2, of the irreducible polynomial X4 9 E Q[X]. Then
< <
DEFINITIOX 4.2.5. A finite extension EIF is said to be strongly GKneser if it is a G-radical extension such that the extension E I K is K*GKneser for every intermediate field K of E / F . The extension EIF is called strongly Kneser if it is strongly G-Kneser 0 for some group G . The concept of strongly G-Kneser extension can be reformulated as follows.
hence I Q ( m ) / ~ 1 = 4 = [Q(I-):Q]. extension. Since Conseq~entlyQ ( Y q ) / Q is a Q (-)-~neser
THEOREM 4.2.6. The following statements are equivalent for a finite G-radical extension E/ F. (1) EIF is strongly G-Kneser. (2) K I F is K * n G-Kneser for every intermediate field K of E / F . (3) E I K is K*G-Kneser for evey intermediate field K of E I F . (4) [ E : K ] = IG/(K* n G)I for every intermediate field K of E I F . (5) [ K : F] = [ ( K *n G ) / F * I for every intermediate field K of E I F .
PROOF. The equivalences (1) Proposition 4.2.1.
e (2) ++
(3) follow immediately from
The implications (2) ==+ (5) sncl (3) ===+ (4) are clear in view of the definitions of the involved concepts. we deduce that i = ( G ) ' / 3 E Q ( I q ) , hence (5) ===+ (2): First, observe that if we take in (5) as K the field E we obtain [ E : F ] = ( ( E *n G ) / F * ( = ( G / F * ( , that is, EIF is a G-Kneser extension. Now let K be an arbitrary intermediate field of the extension EIF. Then, the extension F(K* n G ) / F is K * nG-Kneser by Proposition 2.1.11, SO [ F ( K * n G ) : F] = I(K* n G)/F*I. But [ K : F] = [ ( K *n G)/F*I by hypothesis, and F ( K * n G) C K , hence K = F ( K * n G). Consequently K / F is K * n G-Kneser. (4) (5): If we take in (4) as K the field F we obtain
It follows that we can consider the intermediate field K = Q(P(J6) of the ) /Q. extension Q( We claim t h a t " a g / q ) / ~ is not a K * Q ( a ) - K n e s e r extension, for otherwise, by the proof of Proposition 4.2.1, it would follow that 2 = [Q(I-9)
= I Q (-)/(q
:K
] = [ Q ( Y 3 ) :Q(JS)]
J 6 ) * nQ
(m)) 1 = IQ( G ) / Q
I = 4,
98
Chapter 4
4.2 Strongly G-Kneser extensions
that is, E/F is a G-Kneser extension. Now, for any intermediate field K of E / F , we have which proves the desired implication. Clearly, any strongly G-Kneser extension is G-Kneser, but not conversely, as Example 4.2.4 shows. Now consider an arbitrary G-radical extension E / F . Recall that we have previously introduced the following notation: G={H(F*<H<G), E = { K I F S K , KsubfieldofE}. Clearly, the posets (9, E) and (E, C) are lattices. We have noticed in Section 4.1 that the maps
(5) is G-Kneser, and the maps - n G : E + G, F ( - ) : G + '5 . . EIF . are isomorphisms of lattices, inverse to one another. (6) E/F is a G-Kneser extension with G I F*-Cogalois correspondence. (2): Suppose that E/F is strongly G-Kneser and let PROOF. (1) K E E. Then, the extension K I F is K * n G-Kneser by definition, so
K=F(H)=$(H),withH=KnG=K*nG.
(2) w (3) follows from Proposition 4.1.3 (2). (4): Let K I , K 2 E E be such that cp(K1) = q(K2), i.e., K l n G = (2) K 2n G . M 7 e have K 1 = $(HI ) = F ( H 1 ) and K 2 = $(H2) = F(H2) for some 2 H 1 , H2 E G. Then K l n G = F ( H , ) n G = HI and K 2 n G = F(H,)nG = H by Proposition 2.1.11. Since K1 n G = K 2 n G, it follows that H I = HL, and so K1 = K2. Thus cp is injective. (4) =+ (5): For every H E
G we have
define a Cogalois connection
by Proposition 2.1.11. This shows that cp o = l g , which implies that cp is surjective. Then p is bijective, and $ is its inverse. Consequently, cp and $ are isomorphisms of posets, and so, isomorphisms of lattices inverse t o one another. (5) (1): Let K E E. We are going t o show that K / F is a K* n GKneser extension for every K E E. Since $ is surjective, it follows that K = $ ( H ) = F ( H ) for some H E G, i.e., H = cp(K) = K n G. Since E/F is G-Kneser, the extension F ( H ) / F is H-Kneser by Proposition 2.1.11, i.e., K / F is K * n G-Kneser. the lattice Subgroups ( G I F * ) (5) C J (6): For simplicity, denote by of all subgroups of the quotient group G = G I F * . Since the lattices G are canonically isomorphic, it is clear that everything about the and standard Cogalois connection associated with EIF
which we called the standard Cogalois connection associated with E / F . The next result gives a characterization of G-Kneser extensions E J F for which the standard associated Cogalois connection gives rise t o a bijective correspondence between E and G, and is somewhat dual to the corresponding result for Galois extensions stated in Proposition 4.1.4.
THEOREM 4.2.7. The following assertions are equivalent for a finite G-radzcal extension E/ F. (1) E/F is strongly G-Kneser. (2) E/F is G-Kneser, and the map $ : G + E, d ( H ) = F ( H ) , is surjective. (3) EIF is G-Kneser, and every element of E is a closed element in the standard Cogalois connection associated with EIF. (4) E/F is G-Kneser, and the map cp : E + 9 , p ( K ) = K n G, is injective.
can be expressed equivalently using the Cogalois connection
100
Chapter 4
where g ( K ) = ( K n G ) / F * and $(H/F*) = F ( H ) . So, the maps - n G : E + G and F ( - ) : G -+ E are isomorphisms of lattices, inverse to one another if and only if the maps
extensions is large enough, including important classes of finite extensions: Cogalois extensions, classical Kummer extensions, generalized Kummer extensions, quasi-Kummer extensions. First, let us recall the basic notation we will use in the sequel:
&&+C
and
&&+& are lattice isomorphisms, inverse t o one another, i.e., by Definition 4.1.5, if correspondence. i J and only if E/F is an extension with b - ~ o g a l o i s
The next result shows that the property of an extension being strongly Kneser behaves nicely with respect to subextensions and quotient extensions. where n E N* Recall from Section 3.1 that an extension EIF is called pure when p P ( E ) F for all p~ P. The next definition introduces a concept which can be interpreted as a sort of "local purity". DEFINITION 4.3.1. Let EIF be an arbitrary extension and let n E N*. The extension E/F is called n-pure if p p ( E ) 5 F for every p E Pn. O
1
PROPOSITION 4.2.8. Let EIF be a strongly G-Kneser extension. Then,

for any intermediate field K , the following assertions hold. (1) K I F is strongly K * n G-Kneser. (2) E I K is strongly K'G-Kneser. PROOF. (1) Let L be an intermediate subfield of the extension K I F . Since L E E, there exists H with F* H G and L = F ( H ) by Theorem 4.2.7. But H F ( H ) * = L* K* and H G, hence F* 6 H G n K * . This shows that K / F is strongly K * n G-Kneser.
<
<
< < <
<
(2) For every subfield M of E with K C M E E , the extension E I M is M*G-Kneser since EIF is strongly G-Kneser. But M * G = M * (K*G), so E l M is M * (K*G)-Kneser. It follows that the extension E I K is strongly K * G-Krleser . 0
;
!
I
Clearly, an extension EIF is pure if and only if it is n-pure for every n E N'. Also, note that an n-pure extension is not necessarily pure; e.g., Q( J q ) / Q is Zpure but not pure. Recall that for an arbitrary G-radical extension E I F , which is not necessarily finite, we have defined the standard Cogalois connection
Y '
E
__t
1C(
4.3. G-Cogalois extensions
between the lattices
In this section we introduce the basic concept of G-Cogalois extension, which plays in the Cogalois Theory the same role as that of Galois extension in Galois Theory. A G-Cogalois extension is nothing else than a separable G-Kneser extension with G/F*-Cogalois correspondence. We characterize G-Cogalois extensions EIF within the class of GKneser extensions by means of a certain sort of "local purity" we have called n-purity, where n is the exponent of the finite group G I F * . Using this characterization, we will show in Chapter 7 that the class of G-Cogalois
I = { K 1 F C K , K subfield o f E } ,
G={HIF*<H<G)
a s follows:
p:E+G,
p(K)=KnG,
102
Chapter 4
4.3 G-Cogalois extensioris
103
The characterizations of strongly G-Kneser in terms of this standard Cogalois connection given in Theorem 4.2.7 are strengthened for separable strongly G-Kneser extensions by the next result.
THEOREM 4.3.2 (THEn-PURITYCRITERION). The following assertions are equivalent for a finite separable G-radical extension EIF with G I F * finite and n = exp(G/F*).
(1) E/F is strongly G-Kneser. (2) E/F is G-Kneser, and the map I I , : G -+ E, $(H) = F ( H ) , is surjective. (3) EIF is G-Kneser, and every element of E is a closed element in the standard Cogalois connection associated with E / F . (4) E/F is G-Kneser, and the map cp : E -+ G, cp(K) = K n G, is injective. (5) E/F is G-Kneser, and the maps - n G : E + G, F(-) : G -+ E are isomorphisms of lattices, inverse to one another. (6) EIF is a G-Kneser extension with G I F*-Cogalois correspondence. (7) E/F is n-pure. PROOF. The equivalences (1) through (6) hold for extensions which are not necessarily separable, and are contained in Theorem 4.2.7. (7) =+ (1): Assume that the extension E / F is n-pure, and let K E E. We have to prove that K/F is K * n G-Kneser. For this, we shall use the Kneser Criterion (Theorem 2.2.1). Let p be an odd prime such that Cp E K* n G. If p 1 n , then 6 E F by n-purity. If gcd(p, n) = 1, then there exist a , b E Z with 1 = up bn, hence
(2) ===+ (7) Suppose that the extension EIF is G-Kneser and $ is surjective. Let p be an odd prime with p ) n . There exists then a g E G such that ord(5) = p by Proposition 1.4.8. By Proposition 2.1.10, the extension F ( F *( g ) ) / F is F*(g)-Kneser, hence
Assume that 5, E E. Since XP - gP E F [ X ] and [ F ( g ) : F ] = p, it follows that Min(g, F ) = XP - gp. Then Min(CpglF ) = XP - gp, and so, [ F(Cpg) : F ] = p. Since F(Cpg) E E and $ is surjective, there exists an h E G such that ord(^h) = p and F(Cpg) = F ( h ) = F ( F *(h)), hence Cpg E F ( h ) , and thus Cp E F(g, h). Since the subgroups ( a ) and (^h) of G / F * have both the same order p, they are either equal, or independent ( i . . ( c ) n (^h) = ( 7 ) ) .Observe that if ( i j ) and (^h) are independent subgroups of G I F * , then ( ) = ( ) @ ( ) On the other hand, by Proposition 2.1.11, F ( F *( g, h ) ) / F is F*( g, h)-Kneser, hence
since F C F(Cp) C F(g,h), [F(Cp) : F] p - 1, and [ F ( g , h ) : F] E {p,p2), we deduce that [ F(CP) : F 1 = I , i.e., Cp E F . Suppose that 4 1 n and C4 E E \ F . Then necessarily Char(F) # 2 and G I F * contains an element of order 4, say a . Since F(C4) C E and [ F ( & ) : F] = 2, there exists h E G such that ord(^h) = 2 and F ( & ) = F ( F * ( h ) ) = F ( h ) . It follows that C4 = X + ph, for some X,p E F , hence -1 = X2 + 2Xph + p2h2 and h2 E F. Then necessarily 2Xp = 0. Since C4 @ F we deduce that p # 0, hence X = 0. It follows that C4 = ph, hence C4 E G. Set K = F ( ( 1 + C4)g). Then, using the implication (2) (I), we deduce that E I K is K*G-Kneser. Since 1 C4 = (1 C4)gg-l E K * G , it follows that C4 E K by the Kneser Criterion. But [(1 <4)g]4= -4g4 E F* implies that [ K : F] 6 4 , and so,
<
since G n C F * . I'ow suppose that I + (4 E K*n G. Then necessarily Char(F) # 2, for otherwise, it would follow that C4 = 1, hence 1 C4 = 0 o K* n G, which is a contradiction. Denote nl = ord(1 (4). Since (1 c ~ = ) -4 ~ E F* one has m E {1,2,4). If m = 1 then 1 C4 E F * , hence C4 E F. If m = 2 then (1 C4)2 = 2 6 E F * , hence C4 E F since Char(F) # 2. If rn = 4 then E F by 4 1 n. But 1 C4 E K * n G 2 E implies that C4 E E , hence n-purity. Thus, E/F is strongly G-Kneser. Observe that we did not usi here the separability condition of the extension E / F .
= Xo
+ Xlg -I-XI C4g -I- 2 ~ 2 ~ + 4 2X3(4g3 9~ - 2X3g3

C4
c4
for some Xo, XI , X2, X3 E F . Since C4 E G, we deduce by Corollary 2.1.10 that modulo F* with one of the following elements
must be congruent
104
A
Chapter 4
4 . 4 T h e Kneser group of a G-Cogalois exterisiori
105
But o r d ( 6 ) = 2 and ord(5) = 4, hence necessarily = 7, i.e., G E F*, which contradicts our initial assumption that c4 6 F. This completes the C1 proof of the theorem. In the class of finite extensions EIF with G/F*-Cogalois correspondence the most interesting are those extensions which additionally are separable. In view of the equivalence (1) u (6) in Theorem 4.3.2, these are precisely the separable strongly G-Kneser extensions. They deserve a special name. DEFINITION 4.3.3. An extension E/F is called G-Cogalois if it is a 0 separable strongly G-Kneser extension. REMARKS 4.3.4. (1) The condition " E / F is G-Kneser " cannot be removed from the statement of Theorem 4.3.2 (see Exercise 2). (2) The example in Exercise 2 is an extension with G/Q-Cogalois correspondence, which is not G-Cogalois, but, by Exercise 3, it is 6-cogalois for another group 8. We shall examine such situations more closely in Section 4.5.
<
Recall that if A is a multiplicative group with identity element e? then for any p E P we denote by t p ( ~= ) {XE
A~XP'
= e for somen E
N)
the pprimary component of the group A. Also recall that for any extension E/F we have denoted by Cog(E/F) the torsion subgroup t ( E * / F * ) of the quotient group E * / F * . By Cog2(E/F ) we shall denote the subgroup of Cog(E/F) consisting of all its elements of order 6 2. THEOREM 4.4.1. Let E/F be a separable G-Kneser extension with n = exp (GIF*). (1) Suppose that n f 2 (mod 4). Then EIF is a G- Cogalois extension if and only if
i
(2) Suppose that n if and only if
G/F* =
@ t,(Cog(E/F)). PEP,. 2 (mod 4). Then E/F is a G-Cogalois extension
(3) A strongly G-Kneser extension is not necessarily separable, as the example in Remark 3.1.10 shows. 0
The next result shows that the property of an extension being GCogalois behaves nicely with respect to subextensions and quotient extensions.
Since exp(G/F*) = n , one has tp(G/F*) = 0 for PROOF. (1) all p E P \ P, by Proposition 1.4.7. Therefore, decomposing the finite Abelian group G I F * into the direct sum of its pprimary components, we obtain
PROPOSITION 4.3.5. Let E/F be a G-Cogalois extension. Then, for every intermediate field K of E I F , the following assertions hold. (I) K I F is K * n G-Cogalois. (2) E / K is K*G-Cogalois.
PROOF. The result follows immediately from Proposition 4.2.8.
So, it is sufficient t o show that tP(G/F*) = tp(Cog(E/F>> for every p E P,,. The inclusion " " is clear since G / F * C Cog(E/F). F ' , , and let g E T ( E / F ) with For the opposite inclusion, let p E I 5 E tp(Cog(E/F)). Set G' = G ( g ) and n' = exp (G1/F*). Now observe that P,,I = P,,, and 4 1 n if and only if 4 1 nl, hence E/F is nl-pure since it is 7%-pure by Theorem 4.3.2. But E/F is a separable GI-radical extension with G1/F* finite, hence E/F is GI-Cogalois. Since clearly F* G G' and E = F ( G ) = F(G1), one deduces that G = G', again by Theorem 4.3.2. Consequently, g E G, as desired.
4.4. T h e K n e s e r g r o u p of a G-Cogalois extension
< <
This section essentially shows that a separable G-Kneser extension is GCogalois if and only if the group G has a prescribed structure. This implies that the group G / F * of any G-Cogalois extension EIF is uniquely determined; it is called the Kneser group of E/F and denoted by Kne(E/F).
"*": If G I F * = tp(Cog(E/F)), then we have t o prove that E/F is G-Cogalois, that is, E / F is n-pure. Let p be an odd prime, p 1 n.
106:
Chapter 4
4 . 4 'I'he Kneser group of a G-Cogalois extension
107
Cp
If & E E , then clearly E tp(Cog(E/F)). hence E G I F * , i.e., E G. Using the Kneser Criterion we deduce that Cp E F . If 41n and C4 E E , t h e n l + C 4 E E . I f C h a r ( K ) = 2 , t h e n C4 = 1~ F.
cp
So, we may assume that Char(F) # 2. Then 1 E t2(Cog(E/F)) E G / F * since (1+ C4)4 = -4. Thus 1 C4 E G, SO,by the Kneser Criterion we deduce that [4 E F . This proves that EIF is n-pure, as desired.
+ c4
(2) "=+": Suppose that EIF is a G-Cogalois extension. As in (1) one shows that tp ( G I F * = tp(Cog(EIF)> holds for every odd prime p E P,. Observe that since n r 2 (mod 4), G I F * contains no element of order 4, hence t2(G/F*) is exactly the set of all elernents of G / F * of order 2, and so, t2( G I F * ) C Cog, ( E I F ) . To finish the proof of this implication we only have t o check the opposite inclusion
is a GI-Cogalois extension. Then, using the bijective Cogalois correspondence (provided by Theorem 4.3.2) between the lattice of all intermediate fields of El/F and the lattice of all subgroups of the cyclic group G l / F * of order 4, one deduces that the extension El IF has only one proper intermediate field. On the other hand, one shows similarly that the extension E1/F is H1-Cogalois. Now, using the bijective Cogalois correspondence between the lattice of all intermediate fields of EI/F and the lattice of all subgroups of H I I F * , one deduces that the group H 1 / F * of order 4 is 0 necessarily cyclic, and then, 4 1 n. The next corollary provides a more precise result than that given by Lemma 3.1.6.
<
COROLLARY 4.4.3. Let E/F be a finite G-radical pure separable extension. Then, the extension E / F is simultaneously Cogalois and GCogalois, and Cog ( E I F ) = G I F * .
PROOF. By the Greither-Harrison Criterion, the extension E/F is Cogalois, hence, by Theorem 3.2.3 and Theorem 4.3.2, if follows that EIF is T(E/F)-Cogalois. In particular, it follows that G / F * is finite. On the other, if n = exp (GIF*), then EIF is also n-pure; so, by Theorem 4.3.2, EIF is G-Cogalois. By Corollary 4.4.2, we have G = T ( E / F ) . Hence Cog ( E I F ) = T ( E / F ) / F * = G I F * . Kote that the equality 0 G = T ( E / F ) also follows from Lemma 3.1.6. REMARKS 4.4.4. (1) The condition " EIF is a separable G-Kneser extension" in Theorem 4.4.1 cannot be replaced by the weaker one " EIF is a finite separable G-radical extension with G I F * finite". Indeed, the quadratic G-radical extension Q(c3)/Q, with G = Q+(c3 ), considered in Exercise 2, is not G-Cogalois, but G / p = t3(Cog(Q(&)/Q)) and = 3. I G / V I = exp(G/Q-+)
2 (mod 4), (2) If EIF is a G-Cogalois extension with exp ( G IF * ) 0 then we may have t 2 ( G / F t ) # t2(Cog(E/F)) (see Exercise 5). By Corollary 4.4.2, for any G-Cogalois extension, the group G is uniquely determined. So, it makes sense to introduce the following concept.
Cog2 ( E I F ) C t2 (GIF*). E Cog2(E/F). Set G' = G ( h ) and n' = exp(G1/F*). Then So, let n' f O (mod 4), hence EIF is nt-pure. Now continue as in (1).
"=": If G / F * = tp(Cog(E/F))) @ Cog2(E/F), we have t o prove that EIF is G-Cogalois, that is, EIF is n-pure. Let p be an odd prime in P,. As above, one shows that pp(E) = p p ( F ) . But n 2 (mod 4), hence we cannot have 4 ( n. It follows that the extension E/F is n-pure. 0
COROLLARY 4.4.2. Let EIF be an extension which is simultaneously G-Cogalois and H -Cogalois. Then G = H . PROOF. Set m = exp ( G I F * ) , 71 = exp ( H I F * ) , and k = [ E : F ] . Then IG/F*I = IH/F*I = [ E : F] = k. On the other hand, by Proposition 1.4.10, the order and the exponent of a finite Abelian group have the same prime divisors. Consequently, we P ' , , = Pk, hence by Theorem 4.4.1, it is sufficient t o prove that have P,, = I 41m 41n. Assume that 4 1 m. Then, by Proposition 1.4.8, G I F * contains an element of order 4, say lj. Set G I = F* ( g ) and El = F ( G I ). By TheH, El = F ( H l ) and orem 4.3.2, there exists H I such that F* H I IH1 IF*1 = 4. By Proposition 4.3.5, El / K is an E; flG-Cogalois extension. But E r f l G = E 1 n G = F ( G l ) n G = GI byProposition2.1.1, hence E,/F
<
<
DEFINITION 4.4.5. If EIF is a G-Cogalois extension, then the group G / F * is called the Kneser group of the extension EIF and is denoted by Kne(E I F ) .
108
Chapter 4
Kote that for any extension E I F , Kne(E/F) is a subgroup, in general proper, of the Cogalois group Cog(E/F) of E I F .
implies that G I F * is a finite group, since any infinite group has infinitely 0 many subgroups. The next result shows that the property of a radical extension being with Cogalois correspondence is inherited by subextensions and quotient extensions.
4.5. Almost G-Cogalois extensions

The aim of this section is t o investigate the G-radical extensions with G/F*-Cogalois correspondence. First, we present some general results about such extensions which are not necessarily finite. Then, we focus on finite separable G-radical extensions EIF with G/F*-Cogalois correspondence. We call these extensions almost G-Cogalois. Finally, we characterize those almost G-Cogalois extensions which are not G-Cogalois, but are d - ~ o ~ a l ofor i s a suitable group d . In Section 4.1 we have introduced the concept of extension with G/F*Cogalois correspondence as follows. Let EIF be an arbitrary G-radical extension which is not necessarily finite, and denote
PROPOSITION 4.5.3. Let E/ F be a G-radical extension with G I F* - Cogalois correspondence. Then, for eve? K E I , KIF is an extension with ( K n G ) / F * -Cogalois correspondence, and E I K is an extension with ( K * G ) / K *-Cogalois correspondence.
'
PROOF.Let H = K
n G, and set
R'
3 t
= {HIIF*<H'<H), = {H1'IK*<H''<K*G),
G
&
{HIF*<H<G), = { K IF C_ K , K subfield of E ).
=
S = {SIH<S<G), IC' = { L'I F C L', L' subfield of K ) , IC" = {L1' I K C L", L1' subfieldof E ) .
Now, consider the following order-preserving maps:
We say that EIF is an extension with G/F*-Galois correspondence if the canonical maps v:&-+G, cp(K)=KnG,
$ G -+ &, $(H) = F ( H ) , are bijections, inverse t o one another. In that case, they are actually isomorphisms of lattices. Such extensions, which additionally are separable and have the group G I F * finite, deserve a special name.
-+ K', H' H F ( H 1 ) , p' : IC' -+ 3t1, L' H L ' n H, a'' : R" -+ IC", H" e K(H1'), p'' : IC" + R", L" H L1' n K*G.
a' : 3t'
Since EIF is an extension with G/F*-Cogalois correspondence, then, by definition, the maps
DEFINITION 4.5.1. An extension EIF is said to be almost G-Cogalois if it is a separable G-radical extension with G/F*-Cogalois correspondence, and with G I F* finite. A strictly almost G-Cogalois extension is an almost G-Cogalois extension which is not G-Cogalois. 0 Observe that, by Theorem 4.3.2, a G-Cogalois extension is precisely an allnost G-Cogalois extension which is also G-Kneser. REMARK 4.5.2. The condition "GIF* is a finite group" in Definition 4.5.1 can be replaced by the condition " E I F is a finite extension". Indeed, since E/F is finite and separable, then & is a finite lattice, hence the lattice G, which is isomorphic t o E , also has to be finite. This clearly
are isomorphisms of lattices, inverse t o one another, hence so are also their restrictions t o the sublattices 3t' and IC', respectively, since H = p ( K ) and K = $(H). This shows that the extension K I P is an extension with H/F*-Cogalois correspondence. A similar argument shows that the maps
$" : S --4 IC", S H K ( S ) , 9 ' ': IC" --+ S , L1' L 1 ' n G ,

are isomorphisms of lattices, inverse to one another. Since G I H = G / ( K * n G) E (K*G)/K*,
110
Chapter I
4.5 Almost G-Cogalois extensions
111
the maps
are isomorphisms of lattices, inverse to one another. For every H" E Nf and L" E IC" we have a1'(H") = K(H1') = K((X o p)(H1')) = K ( K * ( H Un G))
= K(H"
for otherwise we would have g = c E F, contradiction. Then, g = [ic E G for some have j = 0 because ord(?j) = p > 1, hence If a # Fp, then XP - a is irreducible [ F ( g ) : F] = p, hence pI [ E : F ] .
cp ci <
i.e., ord(?j) = 1, which is a j, 0 ( j ( p - 1. We cannot E ( ) G , as desired. in F [ X ] by Lemma 1.3.2, so
LEMMA 4.5.6. I f EIF is an extension with G/F*-Cogalois correspondence, then Cq E F whenever 1 Cq E G.
n G ) = ($I1 o p)(HI1)
and 8"(L1') = L" n K * G = (A o p)(LUn K * G ) = K*((L1'n K*G) n G)

= K * (L"
PROOF. Assume that $ ! F. Observe that = 1 E F if Char(F) = (2) in 2. So, necessarily Char(F) # 2. By the proof of implication (1) Theorem 2.2.1, we have
ord(1 On the other hand, F ( F *( 1 + (4 )) = FK4) = F ( F *(C4)). Since ( I ~ 4 =) 2c4 ~ and Char(F) # 2, we deduce that E ( 1 C4 ) G. ( b ) ,F*( 1 + c4) E G, and then, since EIF is an extension with Thus F* G I F*-Cogalois correspondence, we deduce that F*(en) = F* ( 1 c4). This implies that
c4
c4
n G)
+ c 4 ) = I F * ( l + c 4 ) / F * I = 4.
(A 0 v")(L") .
Thus, all = $I1 o p and ,L?"= Xo p". Hence a" and Dl1 are isomorphisms of lattices, inverse to one another, i.e., E / K is an extension with (K*G)/K*Cogalois correspondence. 0 From Proposition 4.5.3 we deduce at once the next result.
c4
+ < +
COROLLARY 4.5.4. Let E/F be an almost G-Cogalois extension. Then, for every K E E, K / F is an almost K n G-Cogalois extension, and E I K is an almost K * G -Cogalois extension. 0
LEMMA 4.5.5. Let E/F be a finite e&ension, and let G be a group such E* and G I F * is finite of exponent n. Then, for every that F* ( G p E 'P,,, one has either C, E G or p l [ E : F ] .
which is a contradiction.
<
PROOF. Let p E P,. By Proposition 1.4.8, there exists g E G such that ord(g^)= p. Then gp = a for some a E F * . = 1 E G, as desired. First, assume that p = 4. If Char(F) = 2, then So, we can suppose that Char(F) # 2. If a E -4F4, then there exists b E F* such that a = -4b4. Then g4 = -4b4 = 4Cb4. Hence 0 = gn - 4<:b4 = (g' - 2C4b')(g2 + 2C b2). This implies that g' = &2hb2, and so, (4 E G since Char(F) # 2. ! - 4 ~ " then X" - a is irreducible in F [ X ] by the VahlenIf a $ Capelli Criterion (Theorem 1.3.5) since a $ F'. Then [ F ( g ) : F] = 4, hence 4 1 [ E : F].
c4
We are now going t o introduce the notation that will be used in the next theorem. Let EIF be a strictly almost G-Cogalois extension, with exp(G/F*) = n. Then, the finite separable extension EIF is not G-Kneser, hence, by the Kneser Criterion and Lemma 4.5.6, there exists a t least an odd prime p with Cp E G \ F . Then ord(&) = p, and so, p i n . Let p l , . . . ,pT be all (Cpi) such that hi E G \ F , and denote Ni = F* distinct odd primes in for every i, 1 i r. Also denote
A
< <
Since INi/F*I = p i for every i, 1 ( i ( r , we have

T
Kow, assume that p E P,,, p # 4. We have gP = a E F*. If a E FP, then there exists c E F such that cP = a = gp. Observe that Char(F) # p,
N I F * =@(N,/F*) i= 1
and lIN/F*I= p l ...:p,.
The symbol " $" stands for "internal direct sum of subgroups" (see 1.1.2).
112
Chapter 4
THEORE 4.5.7. M Let EIF be a strictly almost G-Cogalois extension. Then, with the notation above, there exists a group such that EIF is d - ~ o ~ a l oif i sand only if the following three conditions are satisfied. (1) For every i, 1 i ( r , there exists ai E E* such that F(Cpi) = F ( a i ) , ord(Si) = Ei E P,and Cli E F , i n other words, F(C,, ) I F is a classical Kummer extension of prime exponent li. (2) There exists M E G such that
<
F(S,,)/F does not have any proper intermediate field, hence its degree is a prime number, say li, with li \ p i - 1 for all i, 1 i r . On the other hand, since E/F is B ~ o ~ a l o ithe s , extension F(N,) is X-Kneser, so
< <
[ ~ ( x: F] ) = [ F ( G i ) : F ] = I%/F*
I = li
gcd(IMIF'1, (NIF'I) = 1, and 11 .. el, IM/F*I = [ E : F ] . (3) F ( M ) / M is M-Cogalois, and gcd( ( M I F *I,h . . . 1,) = 1. If the conditions above hold, then
a .
for all i, 1 i ( T . = F * ( a i ) , 1 i 6 r. Then F ( K ) = Let ai E G be such that F ( a i ) , and so [ F ( a i ) : F ] = [ ~ ( x: F] ) = li. This implies that XIi -a: is an irreducible polynomial in F [XI. Since F(Cpi)/F is a Galois extension of degree li, every conjugate of ai over F , in particular Cliaimust belong t o F(CPi) = F ( a i ) , hence CIi E F(Cpi). Then F F ( C i ) C F((,,). Note that F(Cli) # F(5,,) since [F(Cli) : F] li - 1 < [F(C,,) : F] = lie But li is a prime number, hence F((ii) = F , i.e., CLi E F. This proves condition (1).
< -
<
<
PROOF. First, we assume that there exists d such that EIF is d Cogalois. Then F* ( d T ( E / F ) . Denote = {St F* S d}. Since EIF is an extension with both G/F*-correspondence and d / F * correspondence, it follows that for every H E G there exists a unique E such that F ( H ) = F(I?), and the map
Since
is a lattice isomorphism, we have
<
< <
If we set K = F(CP, ,. . . ,Cpp ) , then we have ( N i / F ) of its finite family
is a lattice isomorphism. Let no = e x p ( d / ~ * ) .By Theorem 4.3.2, the extension EIF is nopure. We claim that gcd(no,pi) = 1 for every i, 1 i r. Indeed, if we would have p j [no, i.e., p j E P,, for some j, then Gj E G 2 E , which would imply that Cpj E F by no-purity. But, this would contradict the choice of the primes pi, 1 ( i ( r , being such that Cpi E G \ F . Thus, by Proposition 1.4.10, we also have gcd( 1 ~ / F * Ipi) , = 1 for every i, 1 i ( r . By the definition of the lattice isomorphism T considered above, we have F ( K ) = F ( N i ) = F(Cpi) i r . Since EIF is an almost G-Cogalois extension, for every i, 1 the map 1C, : G + &, $(H) = F ( H ) , is a lattice isomorphism, so, for every i, 1 i r , its restriction t o the lattice { S I F* S Ni ) yields a lattice isomorphism Subgro~ps(N;/F*)-+ Intermediate(F (Ni)I F ) . Since N i / F * is a cyclic group of prime order, it follows that the Abelian extension
< <
Since N I F * is the internal direct sum ( N i lF*)l <i<r of subgroups, it follows that
<
for every i = 1,. . . ,r .-If we transfer this equality into isomorphism T : G --+ G', we deduce that
G via the lattice
< < < <
< <
Since EIF is &cogalois, it follows that K I F is E-cogalois and E I K is ~ * & ~ o ~ a l by o iProposition s 4.3.5, so, in particular K I F is E-Kneser . Thus,
Chapter 4
We are now going to prove that E I K is K*G-Kneser. To do that, we will apply the Kneser Criterion. By Corollary 4.5.4, E I K is an almost K*G-Cogalois extension, so, in view of Lemma 4.5.6, it suffices t o prove that if q is any odd prime with Cq E K*G, i.e., if Cq = a - g for some a E K * and g E G, then Cq E K . Let rn = exp((K*G)/K*). If q 1. m, then 1 = uq vm for some U , v E Z, SO
we deduce that there exists M E G such that gcd(IM/F*I, I N I F * [ )= 1, and lI ...: l,.IM/F*I = [ E : F]. This proves condition (2). We are now going to prove (3). Set L = F ( M ) . Using the fact that the extension El F is an extension with G I F*-Cogalois correspondence, we obtain that To prove that the extension F ( M ) / F is M-Cogalois, we will proceed as above when we showed that the extension E I K is K*G-Kneser. So, it will be sufficient t o check that 6 E F for every odd prime q with C, E M . Let q be such a prime. If q = pi for some i, 1 i r , then Cq = Cpi E M n K L n K = K , as desired. If q $ ! {pl,. . . ,p,), then Cq E F since Cpi E G \ F for every i , 1 i r . Then, the extension L I F is simultaneously Consider MI = T(M) E by Corollary 4.4.2. The M-Cogalois and MI-~ogalois,hence M = internal direct sum decomposition
If q J m , then q J exp(G/F*), hence, by Lemma 4.5.5, we have either or q I [ E : F ] . If Cq E G\ F , then q = p i for some i, 1 6 i r . Thus & = CPi E K . If ql [ E : F ] , then ql JG/F*J since [ E : F ] = J G ~ F * ~ , hence q 1 exp(G/F*). It follows that Cq E F K since EIF is no-pure. Consequently, we have in any case that Cq E K. Therefore, E I K is K*G-Kneser, so also K*G-Cogalois, since it is alno st K*G-Cogalois by Corollary 4.5.4. But, we have already seen that EIK is K'G-~ogalois. Then, by Corollary 4.4.2, we have K * G = K*G. Kow, using the fact that K = F ( N ) = F(N) and the fact that EIF is an extension with both G/F*-correspondence and G/F*-correspondence. we deduce that N = K * n G and f i = ~ * n c .
cq E G
<
< <
< <
c.
MI
Hence
yields via the lattice isomorphism

T:G+g,
Hk-4,
the internal direct sum decomposition
C/F* = ( M I F * ) @ (E/F*).
and so,
i
Finally, we shall prove that ( M / F * (is not divisible by any li, 1 i r . Assume that this is not the case, hence li I JM/F*I for some i. Choose p E M such that o r d ( 3 ) = li. Since the extension F ( M ) / F is M-Cogalois
A
< <
and F ( P ) = F ( F * ( P ) ) , we have [ F ( P ) : F] = I F * ( P ) / F * I = o r d ( P ) = li. = F ( P , 6,). Observe that Let Li= F(P, ai)
Rote that
. .,. . p,) = 1 gcd( ~ G / F * I ,.~

since we have already seen a t the beginning of the proof of this theorem that g c d ( e x p ( G / ~ * )pi) , = 1 for every i, 1 i r . r . Since any Consequently, tpi ( G I F * ) = N i / F * for every i , 1 i finite Abelian group is the internal direct sum of its pprimary components,
F* F * ( p ) n F*(Cp,)
i.e.,
c M nN = F * ,
< <
F * ( p ) n F*(Cp,) = F * . Thus,
< <
F* ( 0 , Cp, ) I F * = (F* ( 0 ) I F * ) 0 (F*( <p,)IF*)
= Zl,x Zp.
Zl,,,
7
116
Chapter 4
4.5 Almost G-Cogalois extensions
117
hence the group F*( P, Cpi ) I F * has exactly four subgroups. Using the lattice isomorphism we deduce that the corresponding extension F ( F *( P, Cpi ) ) I F has exactly four intermediate fields. Kow, observe that Li = F ( F * ( P ,Cpi)), hence the extension L i / F has exactly four intermediate fields. On the other hand, if we transfer the equality F* ( P ) n F*(b,)= F* from the lattice G into the lattice & via the same lattice isomorphism d~: G -+ 8, we obtain that F ( P ) n F ( a i ) = F . But, L i / F is a Galois extension, so (see 1.2.9), By the Fundamental Theorem of Galois Theory we deduce that the extension L i / F has more than four intermediate fields, which contradicts our previous statement. This completes the proof of condition (3). Conversely, assume that conditions (1)-(3) are satisfied, and let the subgroup of E* containing F* such that be
H = F*(Cpi) for some i, 1 i r. Indeed, since Cq E E \ F ( M ) , we have F ( M ) n F ( H ) = F ( M ) n F(C4) = F , hence using the lattice isomorphism $ : G --+ &, S I+ F ( S ) , one finds that ( M / F * ) n ( H / F * ) = {TI. Now, the direct sum decomposition in (2) implies that
hence H / F * 6 t,(M/F*) M / F * in case q 4 {pl, . . . ,p,), which is a contradiction. Thus q = pi for some i , 1 i 6 r , and consequently H/F* t p i ( N / F * ) = F*(Cpi)/F*. Then H = F*(Cpi), hence F(C4) = F(Cpi) = F (ai). Observe that condition (1) implies that [ F ( a i ) : F ] = li. Thus 2 = [ F & ) : F] = [F(Cpi) : F] = li. Since IN/F*I is odd, this implies that li I I M / F * 1, which is a contradiction. If p is an odd prime, choose 4 E M such that ord( = p. Since the extension F ( M ) / M is M-Cogalois, we have
< <
<
<
<
8)
Observe that we have indeed an internal direct sum decomposition in the equality above since G / F * = ( M I F * ) $ ( N I F * ) , N I F * = (Ni/F*), I , : G -+ & is a F ( N i ) = F(Cpi) = F ( a i ) for all i , 1 6 i 6 r , and since I lattice isomorphism. We are going to prove that the extension E J F is e-cogalois. Let no = exp(G/F*). By Theorem 4.3.2, we have only t o prove that the extension E/F is no-pure. Let P E P , , . If p = l i forsome i, 1 < i < r , t h e n Cp=Cl, E F by condition (1). So, we may assume that p $ {I,, . . . ,I,). Then, by (3), we have p l J M / F * ( ,i.e., p J exp(M/F*). Since the extension F ( M ) / F is M Cogalois by condition (3), we deduce that Cp E F whenever Cp E F ( M ) . Kext, we assume that Cp E E \ F ( M ) and we show that we shall obtain a contradiction. If p = 4, then there exists H E G such that F ( 6 ) = F ( H ) since E/F is an extension with G/F*-Cogalois correspondence. Since C4 $ F ( M ) , we have [P(C4) : F ] = 2, hence the extension F(C4)/F has no proper intermediate field. Using the lattice isomorphism cp : E ---+ G, K t-, K nG, we deduce that H / F * is a cyclic group of prime order, say q. We claim that
hence Min(P, F ) = XP - PP. Then, we have also Min(pCp,F ) = XP - pp, and so, [F(PCp) : F] = p. Since F(PCp) E & and E/F is an extension with G/F*-Cogalois correspondence, there exists an H E G such that F ( H ) = F(PCp). Since Cp E E \ F ( M ) and P E M , it follows that F ( M ) n F(PCp) = F . As in the proof of the case p = 4, one deduces that H = F*(Cpj) for some j, 1 6 j r. Therefore, F ( H ) = F(&) = F(Cpj), and then, p = [ F(PCP) : F ] = [F(Cpj) : F ] = lj, which contradict our initial assumption that p 6 { l l , . . . ,1,). Therefore E/F is a e - ~ o ~ a l o i s 0 extension, and we are done.
<
Now, we will apply Theorem 4.5.7 t o nurnber fields. Recall that a number field is nothing else than a subfield K of the field (C of complex numbers such that the extension K / Q is finite.
COROLLARY 4.5.8. Let E be a number field such that the extension
EIQ is
i s some strictly almost G-Cogalois. Then E / Q is b - ~ o ~ a l o for if and only if there exists an intermediate field K of E / Q such group that K / Q is Cogalois, with gcd( ICog(K/Q) J , 6 ) = 1, E = K (C3), and
In this case, we have
118
Chapter 4
PROOF. Suppose that the extension E / Q is 6 k o g a l o i s for some group . Then, the extension E / Q satisfies the conditions (1)-(3) of Theorem 4.5.7. Observe that if p is an odd prime, then Q(Cp)/Q is a classical Kurnmer extension of prime exponent if and only if p = 3. Hence, retaining the notation of Theorem 4.5.7, we have
if we put M = T(K/Q), then conditions (2) and (3) also hold since K / Q is Cogalois and gcd( ICog(K/Q) I, 6) = 1. Finally, observe that Q(C3) = Q ( d q ) with , o r d ( m ) = 2, hence the group E in Theorem 4.6.7 is precisely the subgroup of E* defined by Therefore, the extension E / Q is e - c o g d o i s by Theorem 4.6.7, and we are 0 done.
where Q ( M ) / Q is M-Cogalois, g c d ( l M / q I ,6) = 1, and [ E : Q ] = 2 I M l V I. For simplicity, denote K = Q(M). We shall prove that C, @ K for every p E P. Assume that Cp E K. Then Q C Q(Cp) K . Let H 6 M H be such that Q(Cp) = Q(H). Since the extension Q(Cp)/Q with is H-Kneser, we have I H / q 1 = [Q(Cp) : Q]. If p is an odd prime, then [Q(Cp) : Q ] = p- 1 is an even number, and so is also I M / Q I (since J H / pI divides (M/Q" I), which is a contradiction. If p = 4, then again I M / q I is an even number since [Q(C4) : Q ] = 2, which is a contradiction. Thus, we proved that Cp # K for every p E P. Therefore, the separable M-radical extension K / Q is pure, hence it is Cogalois by the Greither-Harrison Criterion. Then, by Corollary 4.4.3, we necessarily have
v<
EXAMPLES 4.5.9. (1) Let p a positive prime number, and let q = p ' 3 3, r E W .We can choose positive prime numbers t and 1 such tl1a.t
lIqt-1 and g c d ( 1 , q - l ) = g c d ( l , q - 1 ) = 1 . Indeed, let t be any positive prime nurnber which does not divide q - 1, and let d = g ~ d ( l + q + . - - + ~ ~ - 1). ',~1 G t (mod d). Thus, Then q 1 (mod d), so 1+ q + . . . +qt-l 1+ d divides the relatively prime numbers q - 1 and t, hence d = 1. Since qt - 1 = (q - I ) . (1 + q + . . + q t - l ) , surely there exists a prime number 1 such that 1 I qt - 1 and gcd(1, q - 1) = 1. We claim that the extension IFqt / Fq is strictly almost - G-Cogalois, for any group G. wllere G = q ( ), but is not G-~ogalois Indeed, by Fermat's Little Theorem for the field IFqt , the field Fqt contains all the (qt - 1)-th roots of unity of the algebraic closure & of IFq. In particular, since 1 1 qt - 1, we have Cl E IFqt. If we would have Cl E IFq, then C-' = 1, so 1 I q - 1, which is a contradiction. Therefore, (1 E I F q t \IFq, which implies that IFq (Cl) = Fqt since the degree [IFq, : IFq ] = t is a prime nurnber. But t is a divisor of p(1) = 1 - 1, hence t # 1. Let G = T ( G ) Then clearly 4 G < q , . If r n = ord(6), then
Finally, observe that E = Q(G) = Q(M)(Ca) = K(C3) Conversely, let K be an intermediate field of the extension E / Q which satisfies the conditions in the statement of Corollary 4.5.8. We are going t o check that the conditions of Theorem 4.5.7, with F = Q, are fulfilled. First, let p be an odd prime such that Cp E G = T(K/Q)(G). Then C, = yet, with y E T(K/Q) and k = 0,1,2, hence ~ Thus, ~ (r3)P .= 1. If ya = 1, then y E ( b )n T(K/Q) = V , y =6 hence Cp E (C3), and then, necessarily p = 3. If y3 # 1, then we must have K. Being a, y3 E ((,) \ {I), which implies that b E (y3) E T(K/Q) Cogalois extension, K/Q is pure, hence Cp E 0 , which is a contradiction. Consequently we have shown that 3 is the only odd prime in P,, such that Cp E G \ Q, where n = e x p ( G / Q ) , hence N = Q+([a). Now, observe that Q(C3 )/Q a classical Kummer extension of exponent 2 since Q(C3) = Q(J?), so condition (1) of Theorem 4.6.7 is satisfied. Since G / q = Cog(K/Q) @ ( v ( C3 ) / v ), we have C3 @ K , hence [ E : Q ] = [ K (6) : K ] . [ K : Q ] = 2 . [ K : Q ] = 2 ICog(K/Q) 1. Therefore,
rn 1, and SR E q . Hence = 1. Thus 1I m(q - I ) , which implies that 11rn. Consequently, 1 = m, i.e., I G / T I = 1 # t = [Fqt : IFq 1, hence the extension IFqt / IFq is not G-Kneser. Since 1 and t are both prime numbers, the lattices Subgroups(G/T ) and Intermediate(IFqt/IF,) have both only two elements, so the extension IFq* /IFq is almost G-Cogalois. But we have just shown that it is not GKneser, so IF,L/IFq is a strictly almost G-Cogalois extension. We claim q , , the extension IFq. / IFq is not that for any group 21 with G-Cogalois, for otherwise, we would have [Fqt : IFq ] = IG/lI$
<
e('-')
<
t , with
120
Chapter 4
121
t1 : q ) = (qt - l)/(q - 1). Then t 1 (qt - l ) , i.e., qt 1 (mod t). From Fermat's Little Theorem, we have qt q (mod t). Hence q r 1 (mod t), i.e., t ( (q - I). Since gcd(t, q - 1) = 1, it follows that t = 1, which is a contradiction. (2) Let E = ~ ( ( 3 ~ , )F = , Q and G = Q ( G , fi). Then E/F is a strictly almost G-Cogalois extension by Exercise 25. We claim that there Indeed, as in the proof of exists no group 5 such that E J F is 8 - ~ o g a l o i s . Corollary 4.5.8, with the notation of Theorem 4.5.7, we have IN/F*I = 3. Clearly, G / F * = (Q ( a ) / p ) 61 (Q(6)/Q ), hence ( G / F *I = 9. Thus, for any proper subgroup R / F * of G I F * , we must have IR/F*I = 3, and so, condition (2) of Theorem 4.5.7 does not hold. (3) By Exercise 26, there exist infinitely many number fields satisfying the conditions of Corollary 4.5.8. 0
(qt
, , p p ( E ) 5 F for every odd p E P p4(E) F whenever m is even. Prove that EIF is G-Cogalois and G / F * = (Hint: Use Theorem 4.4.1 and Exercise 6.)
(t)
and tp(Cog(E/F)).
8. Prove that condition ( t ) in Exercise 7 is equivalent to condition C2p @ E \ F for every P E Kn. (tt) (Hint: Use the Gay-V6lez Criterion.)
9. Show that a G-Cogalois extension EIF may not satisfy condition ( t t ) in Exercise 8.
10. Let E/F be a G-Cogalois extension. Prove that for any intermediate fields K , L of E/F such that L C_ K , the extension K I L is L*G n K*-Cogalois. 11. Let E/F be an arbitrary extension such that Cog(E/F) is a finite group. If n = exp(Cog(E/F), then show that the extension EIF is pure if and only if it is n-pure. 12. Give an example of a finite separable radical extension which is not G-Cogalois for any group G. 13. (Acosta de Orozco and Ve'Eez [61]). A finite extension E/F is said to have the unique subfield property, abbreviated USP, if for every divisor m of [ E : F] there exists a unique intermediate field K of E / F s u c h that [ K : F ] = m . Let F be any field, and let u E 0 be a root of an irreducible binomial X n - a E F [ X ] with gcd(n,e(F)) = 1, where 0 in an algebraically closed overfield of F . Prove that the extension F(u)/ F has the USP if and only if the following two conditions are satisfied. (a) C, $! F ( u ) \ F for every odd prime divisor p of n. (b) If 4171, then $! F ( u ) \ F.

1. Prove that Proposition 4.1.4 fails for infinite extensions. (Hint: Show that for any field F , the extension F ( X ) / F , where X is an indeterminate, satisfies condition (3) in Proposition 4.1.4, but it is not Galois. Also, any infinite Galois extension satisfies of course condition (1) but not necessarily condition (6) in Proposition 4.1.4.)
2. Show that the G-radical extension Q ( G ) / Q where G = (6), is an extension with G/e-Cogalois correspondence which is not G-Cogalois.
3. Show that the extension Q(c3)/Q is Cogalois.
( a ) - C o g a l o i s , but is not
I p .
4. Show that any finite separable G-radical extension, with exp(G) = 2 is G-Cogalois. 5. Prove that the extension Q(i, &)/Q is Q (i, &)-Cogalois, but + i l l 2 E t2(Cog(Q(i, J Q l Q ) ) \ t2(Q (i, &)/V). 6. Let E/F be a G-Cogalois extension with exp ( G I F * ) z 2 (mod 4). Prove that if p4 ( E ) C F , then t2(G/F*) = t2(Cog(E/F)). (Hint: Proceed as in the proof of Theorem 4.4.1 (I).)
c4
14. Show that a finite extension E/F of degree n has the USP if and only if the canonical map Intermediate(E/F) is a lattice isomorphism.
+ Dn, K
H [ K :F],
15. Prove that the following assertions are equivalent for a finite GCogalois extension EIF of degree n. (a) E/F has the USP. (b) The Kneser group G / F * of EIF is cyclic.
122
Chapter 4
4.7 Ribliographical corrirrierits to Chapter 4
123
(c) G / F *
Z,,.
16. Prove that the following assertions are equivalent for a finite Cogalois extension E/F of degree n. (a) E/F has the USP. (b) The Cogalois group Cog(E/F) of EIF is cyclic. , . (c) Cog(E/F) E Z 17. Let F be any field, and let u E R be a root of an irreducible binomial X n - a E F [ X ] with gcd(n, e ( F ) ) = 1, where R in an algebraically closed overfield of F . Prove that the following assertions are equivalent. (a) The extension F ( u ) / F has the USP. (b) The extension F ( u ) / F is n-pure. (u)-Cogalois. (c) The extension F ( u ) / F is F* (d) The extension F ( u ) / F is G-Cogalois for some group G, and G I F * is a cyclic group. 18. Prove the following statements. (a) Any extension of degree a prime number has the USP. (b) A finite G-radical extension having the USP is not necessarily G-Cogalois. (c) A fin& G-Cogalois extension may fail to have the USP. 19. Show that the condition " G / F * is a cyclic group" in the statement (d) of Exercise 17 is essential. (Hznt: Take F = Q and u = fi ( l + i ) E @. Then u is a root of the irreducible binomial X4 + 16 E F[X], and the extension F ( u ) / F is Q ( i ,4)-Cogalois, but it does not have the USP.) 20. Let u R be a root of an irreducible binomial X" - a E F[X]. If n $ 0 (mod 4) and gcd(n,e(F)) = 1, then prove that the following statements are equivalent. (a) The extension F ( u ) / F has the USP. (b) The extension F ( u ) /F is F*(u)-Cogalois. (c) The extension F ( u ) / F is G-Cogalois for some group G. 21. Let EIF be a G-Cogalois extension, with F a field of characteristic # 2, E = F ( u ) , and u a root in R of an irreducible binomial X" - a 6 F[X]. Assume that the Kneser group G I F * of EIF is noncyclic. Prove the following statements. (a) exp(G/F*) = 2. (b) E/F is a classical 2-Kummer extension.
(c) C2" E E and C4 @ F . (d) There exist s - 1 elements a 2 , .. . ,a, E F* such that E = F(C4,& ,... ,&) and G / F * =F*(C4,Jaz,... , & ) I F * 2 (Z,)". 22. Show that for any field F of characteristic # 2 such that C4 # F, and for a root & E 0 of a polynomial X 2 - a E F [ X ] such that fi # F*((;), the extension F (c4,fi)/ F is a simple radical quartic F*((4, &)-Cogalois extension with a noncyclic Kneser group. 23. Prove that any cyclic Galois extension E / Q of degree G-Cogalois, but has the USP.
> 2 is not
24. Let K be an intermediate field of an extension E/F such that the extension K / F is almost K n G-Cogalois and the extension E I K is almost K*G-Cogalois. Does it follow that the extension E/ F is almost G-Cogalois? 25. (Lam-Estrada, Barrera-Mora, and Villa-Salvador [79]). Show that the extension Q(C3, @ ) / Q is strictly almost G-Cogalois, where G=v(&,~). 26. (Lam-Estrada, Barrera-Mora, and Villa-Salvador [79]). Let a , p E P,p 2 5, a n d l e t F = Q, E =Q(C3,$), and G = Q ( C 3 , f i ) . Show that the extension E / Q is strictly almost G-Cogalois and B/Q = (Q( - ) / Q ) CD ( Q ( @)/Q). 8 - ~ o g a l o i swhere ,
Section 4.1. The concept of Galois connection is discussed in Stenstrom [ l o o , Chapter 111, Section 81 for complete lattices, and in Albu and N&t&escu [18, Section 31 for arbitrary posets. The dual concept of Cogalois connection was introduced by Albu and Nicolae [19]. The nice idea t o use the "prime" notation in the standard Galois connection associated with a field extensions goes back t o Kaplansky [74, Section 31. Proposition 4.1.4, inspired by the nice approach of the Fundamental Theorem of Galois Theory in Kaplansky [74, Theorem 101, seems t o be new. This section basically follows the outline in Albu and Niistikescu [18, Section 31 and Albu and Nicolae [19], although there are several extensions
124
Chapter 4
and improvements.
Section 4.2. The concept of strongly G-Kneser extension is due t o Albu and Nicolae [19]. The results of this section are taken from Albu and Nicolae [19] and Albu and Tena [25]. Section 4.3. The concept of G-Cogalois extension, as well as its basic properties including the n-Purity Criterion (Theorem 4.3.2) are due to Albu and Kicolae [19]. This criterion is the main tool in the whole Cogalois Theory. It permits t o reobtain in a simple and unifying manner, even in a more general setting, a series of results of Albu [3], Barrera-Mora, Rxedowski-Calderbn, and Villa-Salvador [30], Greither and Harrison [63], as well as a considerable part of the classical Finite Kummer Theory (see Chapter 7). A very elementary approach, a t the undergraduate level, t o the basic concepts and properties of Kneser and G-Cogalois extensions is available in Albu 141, [5]. This section basically follows the presentation in Albu and Kicolae [I91 and Albu and Tena [25], although there are several extensions and improvements. Section 4.4. The structure of the group G I F * of a G-Cogalois extension E / F and its uniqueness were established by Albu and Kicolae [19]. The term of Kneser group of a G-Cogalois extension appeared first in Albu and Nicolae [22].
*.
CHAPTER 5
GALOIS G-COGALOIS EXTENSIONS

The purpose of this chapter is to study finite extensions which are simultaneously Galois and G-Cogalois. In Section 5.1 we characterize Gradical extensions, not necessarily finite. which are separable or Galois. In the next section we prove that the Kneser group and the Galois group of any finite Abelian G-Cogalois extension are isomorphic, but not in a canonical way. Section 5.3 contains some applications of Section 5.1 and Section 5.2 t o elementary Field Arithmetic. Further applications, involving results we will prove in Section 6.1, Chapter 7, and Section 8.1 will be given in Section 8.2.
Section 4.5. An analogue of Theorem 5.3 of Greither and Pareigis [64] prompted Albu and Nicolae t o raise in [19] the following problem: if E/F is a separable G-radical extension which is not G-Cogalois, but is an extension with G/F*-Cogalois correspondence, then does there exist another This ? problem was solved in negagroup 5 such that E I F is C - ~ o ~ a l o i s tive by Lam-Estrada, Barrera-Mora, and Villa-Salvador in [79, Section 41. They introduced the concept of a pseudo G-Cogalois extension, which is precisely our concept of finite separable G-radical extension with G / F * Cogalois correspondence. This section mainly presents with some improvements, simplifications and corrections of several inaccuracies, the negative answer of Lam-Estrada, Barrera-Mora, and Villa-Salvador [79] t o the question raised by Albu and Kicolae [19, Problem 3.101 mentioned above.
5.1. Galois G-radical extensions In this section we characterize G-radical extensions, not necessarily finite, which are separable or Galois. Recall that for any torsion group T with identity element e we introduced in Section 1.4 the notation:
r.
.Vi
r
i
OT={ord(x)(x E T ) . of N is a bounded set, or equivalently, a finite set, When the subset one says that the torsion group T is a group of bounded order, and the least number n E W* with the property that T n = {e) is the exponent exp(T) of T . The group T is said t o be n-bounded if T is a group of bounded order and exp(T) = n. Observe that for any G-radical extension E I F , which is not necessarily finite, the group G I F * is a torsion Abelian group, so it makes sense t o consider the set of natural numbers. The next concept is important for infinite extensions; we will deal with it much more deeply in Chapters 12 and 13.
126
Chapter 5
5.1 Galois G-radical exter~sioris
127
DEFINITION 5.1.1. A G-radical extension E I F , which is not necessarily finite, is said to be a bounded extension if G / F * is a group of bounded order; in this case, if exp(G/F*) = n, we say that E/F is an n-bounded extension. 0 If EIF is an n-bounded extension, then, in view of Remark 1.4.9 we have OGIF* = mn. LEMMA 5.1.2. Let E/F be a G-radical extension which is not necessarily finite. Then E/F is separable if and only if gcd (m, e (F))= 1 for all m E OGIF*. PROOF. We can suppose that Char(F) = p > 0, so e ( F ) = p. Assume that E/F is separable, and let m E OGIF*. Then, there exists u E G such that m = ord(G). If p 1 m then m = ps for some s E W . It follows that us is simultaneously separable and purely inseparable over F , so us E F. This shows that G ' = 7, contrary t o ord(2) = m. Conversely, let u f G and set m = ord(2). By hypothesis (p,m) = 1, hence the derivative of the polynomial f, = X m - unl E F [ X ] is nonzero. This means that f, has only simple roots in R. But, the minimal polynomial of u over F is a divisor of f,, hence u is separable over F . Since E = F ( G ) we deduce that E/F is a separable extension. 0 COROI~IARY 5.1.3. Let E/F be an n-bounded G-radical extension, which is not necessarily finite. Then E/F is a separable extension if and only if gcd(n, e ( F ) ) = 1.
= IDr1 for PROOF. Apply Lemma 5.1.2, and use the fact that OGIF.* any n-bounded G-radical extension E/ F . 0
. . . . .p17., that m 2 2. By Lemma 5.1.2 we have (m, e ( F ) ) = 1. Let m = where pl, . . . ,p, are mutually distinct positive primes and kl , . . . ,k, E N* . If we set qi = p k. i a ,i = 1 , . . . ,r , then clearly qi E O G I p for a11 i, 1 i T, hence qi = ord(Ci) for some ui E G, i = 1 , . . . ,T. We claim that Cq, E E for all i, 1 i r. Indeed, set ai = upi and observe that ui is a root of the polynomial fi = XQ. - ai E F[X]. Since E/F is normal, E contains all the conjugates of ui over F . Any conjugate of ui over F is necessarily a root of fi, hence all these conjugates are . .,, . u ~ <where ~ , t l , . . . , are certain qi-th roots of unity. Suppose u~<~ that no is a primitive qi-th roots of unity. Then every cj, 1 j s, will be necessarily a qi/pi-th root of unity. Now, observe that all the conjugates of the element u f i l ~ E ' E over F are
pt'
< <
< <
cj
cs
< <
Thus, ufiIp' coincides with all its conjugates, hence ufiIp' E F since E/F is a Galois extension. But this is a contradiction since qi is minimum with the property that u:' E F * . This proves that there exists j , 1 6 j s, such that tj is a primitive qi-th root of unity. Since ui, uitj E E , we deduce that E E , and so, E E . This proves our claim. E E . For every i, 1 i r , set We are now going t o show that bi = m l p f i . Since gcd(bl, . . . ,b,) = 1, we have 1 = bici for some cl, . . . ,c, E Z. It follows that = ~l+...+br.~v = (c:;)" . . . . . (:;)"' E E,
cj
cqi
ern
x16i<r
< <
c~~~ ~2
since
cq, E E and Cf;f is a primitive qi-th root of unity for all i, 1 < i < r .
An immediate consequence of Corollary 5.1.3 is the following result. COROLLARY 5.1.4. Let E/F be a finite G-radical extension with G / F * finite, and let n = exp(G/Ft). Then E/F is separable if and only if gcd(n?e(F))= 1. 0
l L += " Assume that ( m , e ( F ) ) = 1 and Cm E E for all m E OGIF*. By Lemma 5.1.2, E/F is a separable extension. To prove that EIF is a normal extension, let u E G and set m = ord(u). As above, every conjugate of u over F is a product of u by a power of em, so it belongs 0 t o E . Consequently EIF is a normal extension, and we are done.
COROLLARY 5.1.6. Let E/F be an n-bounded G-radical extension. E E. Then E/F is a Galois extension if and only if gcd(n, e ( F ) ) = 1 and
en
PROPOSITION 5.1.5. Let E/F be a G -radical extension, which is not necessarily finite. Then E/F is a Galois extension if and only if gcd (m, e ( F ) ) = 1 and E E for all m E OCIF*.
PROOF. " " Assume that E/F is a Galois extension, that is, E/F is a separable and normal extension, and let m E OGIE.*. We can assume
PROOF. Apply Propositiorl 5.1.5, and use the fact that OGIF* = IDn 0 for any n-bounded G-radical extension E/ F . COROLLARY 5.1.7. Let EIF be a finite G-radical extension with G / F * a finite group of exponent n . Then EIF is a Galois extension if and only CI E E. if gcd(n, e ( F ) ) = 1 and
c,
130
Chapter 5
5.3 Applications t o elementary Field Arithmetic ( I )
131
PROOF.The extension E/F is T (E/F)-Kneser by Examples 2.1.15 (2), hence it is T(E/F)-Cogalois by Theorem 3.2.3 and Theorem 4.3.2. Thus, Gal(E/F) is isomorphic to Kne(E/F) = T ( E / F ) / F * = Cog(E/F) by 0 Theorem 5.2.2. R.EMARK 5.2.4. An alternative proof of Theorem 5.2.2, involves the concept of lattice-isomorphism of groups (see Section 15.2). What follows is a brief description of this approach which will be presented in full detail in Section 15.3 for arbitrary (i.e., not necessarily finite) extensions. If EIF is a finite Abelian G-Cogalois extension with Galois group I?, then the lattice anti-isomorphism from the lattice Intermediate (E/F ) onto the lattice Subgroups( I? ) given by the Fundamental Theorern of Finite Galois Theory, produces, by taking the characters, a lattice isomorphism from the lattice ')) of all subgroups Intermediate ( E I F ) onto the lattice Subgroups(Ch ( I of I?. of the character group Ch ( I ? ) = Hom(I', C ) On the other hand, since EIF is a G-Cogalois extension, the map HIF* I+ F ( H ) is an isomorphism from the lattice Subgroups (Kne (EIF)) onto the lattice Intermediate ( E I F ) . If we compose these two lattice isomorphisms, we obtain a lattice isomorphism from Subgroups (Kne ( E lF ) ) onto Subgroups (Ch ( I ? ) ) . According to the terminology of Section 15.2, this means that the groups Kne ( E I F ) and Ch ( r ) are lattice-isomorphic. In general, two lattice-isomorphic groups A and B are not necessarily isornorphic, excepting few cases, when the groups A and B satisfy various restrictive conditions, in view of a series of quite technical results due to Baer [27] (see Section 15.2). By chance, some of these conditions are satisfied in our case, and so, one deduces that the finite Abelian groups Kne ( E / F ) and Ch ( I ? ) are isomorphic. Since any finite Abelian group is isomorphic, but not in a canonical way, t o its character group, it follows that the groups Gal ( E I F ) and Kne ( E lF ) are isomorphic. This approach, which is unavoidable for infinite extensions, is more natural than the inductive one provided in the proof of Theorem 5.2.2, but it involves a certain amount of technical facts related to lattice-isomorphisms 0 of groups.
Field Arithmetic, based on results we will establish in subsequent chapters, will be presented in Section 8.2. In the sequel, the algebraically closed field R containing any algebraic number field K is the field C of complex numbers, and, for any n E N*, the primitive n-th root of unity is C,, = cos(2rln) + i sin(2rln). The next result is very similar t o Proposition 3.2.6, but in contrast t o that, it provides a quartic Gdois extension. P~orosr~ro 5.3.1. x The following assertions hold. ~ ( 4 2 is ) a subfield of the field Q( d m ) .
[~(Jcfi) : Q(dZ)] = 2 and

Q(~=)/Q
[Q(m) Q]
:
= 4.
is a n Abelian extension. ~ ( d m ) /4 5 ~) and ( Q ( d Z ) / Q are both Cogalois extensions. Q(Jm)/Q is neither a radical extension, nor a Kneser extension, nor a Cogalois extension.
cog
(Q(dm)/Q = {T, ) 3). The element d r f i of the group Q ( d F f i */ )
( @ '
has infinite
order.
PROOF. For simplicity, set
As in the proof of Proposition 3.2.6, the assertions (a),(b) and (d) are easily shown. (c) Observe that Min(0, Q) = X 4 - 4X2 over Q are
+ 2 hence the conjugates of
and all of them belong t o E, since
5.3. Applications to elementary Field Arithmetic (I)

The aim of this section is t o apply the results of Sections 5.1 and 5.2 to elementary arithmetic of fields. Further applications of Cogalois Theory t o
This shows that EIF is a quartic Galois extension. It is easily shown that EIF has a cyclic Galois group generated by the automorphism sending to d m .
d m
132
Chapter 5
5.3 Applications to elementary Field Arithrrletic ( I )
133
(e) As in the proof of Proposition 3.2.6, it suffices t o show that E/F is not a Cogalois extension. Assume that this is not the case, and look for a contradiction. By Corollary 5.2.3, the assumption that EIF is a Cogalois extension implies that Cog ( E I F ) Z Gal ( E I F ) N Z4. Then clearly exp(Cog ( E I F ) = 4, so i = C4 E E by Corollary 5.1.7, which is a contradiction. This proves that E/F is not a Cogalois extension. For (f) and (g) proceed as in the proof of Proposition 3.2.6.
In the sequel, the notation x f y = z (resp. x f y E M ) , where x, y , z E C and M 5 C, means that either x y = z or x - y = z (resp. either x y E M or x - y E M). Also, the notation x f y # M means that x + y # M and x - y # M .
Q+
R.EMA RK 5.3.2. More generally, consider the extension ET / F , where F = Q, E, = ()(Or), and
PROPOSITION 5.3.3. The following statements are equivalent for d E andre%. (1) The polynomial x4- 2rX2 + r2 - d is reducible in Q[X]. (2) There exist c, k E (& such that " r d = c2 and r f c = k2/2. (3) There exists c E (& such that r2 - d = c2 and
\@
r radicals
We claim that for every r E N,r 2 2, the extension E,/F is not a radical extension. To do that, observe that and Er/F is a Galois extension with Galois group the cyclic group of order 2r generated by the automorphism sending c0s(n/2~+') t o c o s ( 5 ~ / 2 ~ + ' ) (see Exercises 19 and 20). Similar arguments t o those used above prove that if Er/F would be a radical extension, then it would be a Cogalois extension, and Cog (E,/F)
N Gal
PROOF. Set f = x4- 2rX2 roots of f are
+ r2 - d, and observe that the complex

++ (4).
and none of them is rational. Hence (1)
(2): Assume that f is reducible in Q[X]. Since r f E R\Q, (1) it follows that f has necessarily a monic divisor of degree 2 of type
( x * ) .(.*JX)
(E,/F)
N Z2-.
This would imply that Czv E ET2 R by Corollary 5.1.7, which is a contradiction. This proves our claim. 0 The examples discussed in Proposition 3.2.6 and Proposition 5.3.1 have
. For simplicity, denote
hence r2-d=Z On the other hand, forsome

C E ~ .
for any r, d E Q. Kext, we are interested to investigate similar questions for the extension Qr,d /Q, namely the following ones: When is Q,,d / Q a quartic extension? When is Q,,d/Q a Galois extension? When is Qrld/Q a Cogalois, or Kneser, or radical extension?
Notice that we have r > c > 0 and r > > 0, hence all quadratic radicals appearing in the two formulas just above are positive real numbers. Thus,
134
Chapter 5
5.3 Applications t o elementary Field Arithrr~etic (I)
135
This implies that r k c = k2/2 for some k E (2). cation (1)
(&, which proves the impliused in the proof of the implication (1)==+ (2) in Proposition 5.3.3 we will have the following ones
(2) ==+ (3): If r2 - d = c2, then, by the proof of the implication (I) (2), we deduce that
hence , ~4 are suitable numbers in (-1, I ) , depending on the where c ' l , ~ 2ig, chosen quadratic roots and satisfying the relation ~ 1 ~ 2 = ~ -1. 3 ~ 4 This implies that
(3)
(4): Assurne that r2 - d = 2 and
To prove (4), it is sufficient t o show that wise. Then r - c = 2s2 for some s E Q+, hence
dm
6Q.
Suppose other-
Now, continue as in the proof of Proposition 5.3.3.
COROLLARY 5.3.5. For any square-free integer d E Z \ (1) and any n E Z , the polynomial X 4 - 2nX2 n2 - d is irreducible in Q[X], and so, [ Q n , d : Q] = 4.
and so, Thus, there exists t E Q+ such that r r - c = 2s2 we deduce that hence
~ = ~ ( d 0 =/Q r2 ,d ) + c = 2t2. Now, using the equality

r = s 2 + t 2 and c = t 2 - s 2 ,
=Q(m).
PROOF.Assume that the polynomial x 4 - 2 n x % n 2 - d is reducible in ' Q[X]. Then, by Remarks 5.3.4, there exist c, k Q$ such that n2 - d = 6 and n f- c = k2/2. But n2 - d E Z , hence necessarily c, k E W . Thus, 2(n k c) = k"mp1ies that k = 2m for some m E N*, and then, n k c = 2m2. To make a choice, assume that n - c = 2m2. Then n + C = 2(m2 + c ) , and so, which contradicts the fact that d is a square-free rational integer.
" 6 ' = (s2 + t"2 - (t2 - s2)2 = (2st)2 E Q2, d =r which is a contradiction.
Q\
REMARKS 5.3.4. (1) It is easily seen that for r = 0 and for any d E @, the polynomial X 4 - 2rX2 + r2 - d is irreducible in Q[X].
(2) If we do not assume that the rational numbers r and d in Proposition 5.3.3 are positive, then we have t o be careful with the meaning of the normal radicals and appearing in its proof. We are going t o show that Proposition 5.3.3 holds for any r E Q and ; we denote, any d E Q \ @. Indeed. for any complex number a E C \ R as usually, by fi one of the not specified complex roots of the equation x2 - a = 0. Then, instead of the equalities
J r &
Kow, we are going to investigate the existence of quadratic subfields of a1gebra.i~ number fields of degree 4. Let E be an algebraic number field of degree 4, that is, a subfield of the field C of all complex numbers such that i E : Q ] = 4. Such fields are usually called quartic fields. Then, two cases arise: Case 1: The only subfields of E are the trivial ones, i.e., Q and E , in other words the field E has no quadratic subfield. According t o Exercise 5, if n E Z and J is any of the four complex roots of the polynomial X L n X - 1, then the field Q(J) is a quartic field having only two subfields if and only if n E Z \ { -4, 0, 4 ). This is a consequence of a more general result establishing when an arbitrary separable quartic extension E/F has no proper subextension (see Exercise 6).
136
Chapter 5
5.3 Applications t o elerrlentary Field Arithmetic ( I )
Case 2: The extension E has a t least a subfield K with Q # K # E , which has to be necessarily a quadratic field. Then, there exists a squarefree rational integer d # 1 such that K = ~ ( d d )It . is easily seen that there exist a , b E Z such that
This implies that there exist ao, a l ,
a, a3 E Q
such that
We can write (6) as
Since [ E : Q ] = 4 we deduce that b # 0. We ignore whether or not E can be expressed as Q - (. J a ) for a suitable r E Q. However, we can write E = Q ( J w ) , where r E Q and s E Q \ Q 2 . Next, we shall examine this case.
Squaring (7) we obtain Then necessarily a, a1 Jd = 0, for otherwise, (8) would imply that B E Q(Jd), which is a contradiction. Consequently, (7) becomes
PRO PO~ITION 5.3.6. The following statements hold for a quartic alge), where r E Q and d E Q \ @. braic number field Qr,d = Q (a) Qr,d has at least two distinct quadratic subfields if and only if Qrqd has precisely three distinct quadratic subfields if and only if r " dE
( J a
B = &(a, (9) From (9) we deduce that
+ as&).
Q2
(b) Qr,d has precisely one quadratic subfield if and only if r2 - d
# Q2.
PROOF. (1) If r2 - d = c2 for some c E Q, then as in Remarks 5.3.4, we have
Jr+Jd
=*dm+ ,-J
hence = ~ ( J(r f c)/2 k d m ) . Since [ Q r t d : Q] = 4, we deduce that )P? ' FJ , ~ / ) c+ Qr,d ~ ( J (Q = and Q#Q(J(~+c)/~ # )Q ( J ( ~ - c ) / ~#)Q , so Qr,d has at least two distinct quadratic subfields. Since Qrld/Q is in fact a Galois extension with a Galois group G of order 4 and exponent 2, it follows that G is isomorphic t o the group Zz x Z2. Hence G has exactly three subgroups of index (or order) 2, and consequently, by Galois Theory, Qr,d has exactly three distinct quadratic subfields. Conversely, assume that Q r l d has at least two distinct quadratic sub. have t o show that r' - d E @ . fields, one of them surely being ~ ( d d ) We Let Q ( d k ) be another quadratic subfield of Qr,(l, with k E Q \ @ and Q(dk) # Q ( d d ). Then, we have
Thus, u coincides with its conjugate in the quadratic extension Q(dd)/Q, i.e.,
r+dd After easy calculations we obtain
r -dd
a': - 2anapr dug = 0. (10) We have a3 # 0, for otherwise, (10) would imply a2 = 0, and then, (9) would become 6' = 0: which is impossible. Hence, we can divide (10) by a: # 0. If we set t := az/a3, then we see that t E Q is a root of the quadratic equation
(11) Since the roots of (11) are
t2 - 2rt
+ d = 0.
we deduce that r2 - d E
v ,and we are done.
138
Chapter 5
139
(b) By negation, statement (a) can be obviously reformulated as follows: Qr,d has at most one quadratic subfield if and only if r2 - d @ Q 2 . But Qr,d has a t least one quadratic subfield, namely Q ( J 2 ) . This clearly implies assertion (b). 0 Next, we are going t o study when is Qrld/Q, r E Q, d E Q Galois extension.
Case 2: d F d E Q ( d 2 ) \ Q . Then Q r l d has precisely one quadratic subfield by Proposition 5.3.6 (b), hence, by Galois Theory, Gal ( Q r l d / Q ) is a group of order 4 having only one subgroup of order 2. Therefore, this group is necessarily a cyclic group of order 4, and so,
\ @, a
v.
Now observe that
m E Q(&)
\ Q Q d ' F d = s&
for some s E
PROPOSITION 5.3.7. The following statements hold for r E Q and d E Q\@. f the polynomial X 4 - 2rX2 + r 2 - d is reducible i n Q[X], then (1) I Qr,,l/Q is a Galois extension, and Gal (Qr,d/Q) Z2. (2) If the polynomial X 4 - 2 r X 2 + r2 - d is irreducible in Q [XI, then Qr,d/Q is a Galois extension if and only if d F d E Q(& ) . I n this case, we have (a) Gal (Qr,d/Q) % Z2 x Z2 d E p, d m = S& for some s E @ . (b) Gal (Qr,d/Q) Z4
Case 3: E Qr,* \ Q( &). Then Q(&) is a proper subfield of the field Q( dd ) ( d m ) , which in turn, is a subfield of Q , d . Thus
This clearly implies that
PROOF. (1) follows immediately from Proposition 5.3.3, since any quadratic extension is a Galois extension. (2) Set f = x4- 2 r X 2 of f are
hence has a t least two quadratic subfields, and then, by Proposition 5.3.6 (a) we must have h2-d E Q, which contradicts our assumption 4 E Qr,d \ ~ ( d d ) This . shows that Case (3) cannot occur, and n proves the proposition.
Y
+ r2 - d and observe that the complex roots
COROLLARY 5.3.8. The following assertions are equivalent for a n n E Z and a square-free integer d E Z \ {I). (1) Q n l d / Q is a Galois extension. (2) There exists k E Z* such that either
d m=k
or
m=
hence f = in ( J a , 0 ) . Since
k&.
F d ,/'a7
Jr-Jd=* J
we have The following cases arise:
(3) n2 - d E W 2 , or there exist x, y E Z* such that x2 - dy2 = -1 and n = dy.
PROOF. By Corollary 5.3.5, the polynomial f = X" 2nX2 + n2 - d is irreducible. Then, by Proposition 5.3.7 (2), QTlld / Q is a Galois extension m E Q ( d d ) . There are two cases. if and only if d
Case 1:
4 -
Case 1: d F d E Q. Then, the proof of Proposition 5.3.6 (a) shows

that
Qr,d = Q ( J ( r + c ) I ~ , where c E C& is such that r2 - d = c", and Qr,d/Q is a Galois extension with Gal (Qr,d /Q) Z 2 x Z2.
Case 2: d x, z E Q. This irnplies that
p.Then we have d m Q( J d ) \ Q. Then d

E
d E Z*.
m = z + x&
for some
Since { 1, A ) is linearly independent over Q, we deduce that s z = 0. But x # 0 since d m @ Q, hence necessarily z = 0, and then d m=
140
Chapter 5
5.3 Applications to elerrieritary Field Arithmetic ( I )
141
x&, or equivalently d(n2 - d) = ( ~ d ) ~ . We deduce that c := xd E Z and d 1 c2. Then d I c since d is square-free, hence x E Z . Thus n2 - d = dx2. Then d I n2, and consequently d ( n since d is a square-free integer. So n = dy for some y E Z * verifying the Pell-Fermat equation Conversely, if (x, y) E Z * x Z * is any solution of the Pell-Fermat equation above, and n = dy, then dn"-dn"-d = k x a , and we are done. 0 We are now going to find conditions on the rational numbers r, d for which Qr,d/Q is a Cogalois extension. LEMMA5.3.9. Let d E Q\Q2. Then, Q(J;i)/Q if and only if -d # @ and -3d # p .
is a Cogalois extension
In view of the Greither-Harrison Criterion (Theorem 3.1.7), we have t o prove that Q(J&, & ) / Q is a pure extension. So, let p E P be such that Cp E E. Then Q(C,) E E , hence p - 1 = [Q(Cp) : Q ] 4, and so, P E (374, 5). We cannot have p = 5, for otherwise, it would foilow that E = Q(C5), hence Z42 Gal(Q(C5)/Q) = Gal(E/Q) 2 Z2 x Z2, which is a contradiction. If p = 4, then Q(C4) = Q(i) has t o be one of the three quadratic subfields Q(J&), Q(J&), and Q(J&&) of E. For instance, if we would have Q(i) = Q(J&'), this would imply that i = a& for some a E q , i.e., -dl E Q 2 , which contradicts our assumption. Similarly, if p = 3, then Q(&) = Q(J-3) has t o be one of the three quadratic subfields Q(J& ) , Q(J& ) , and Q ( m ) of E , and this cannot happen either. Consequently Cp # E for every p E P. This shows that E / Q is a pure 0 extension, hence a Cogalois extension.
<
THEOREM 5.3.11. The following statements hold for r E Q and d E
PROOF. For the given d E Q \ Q2 we can find a unique square-free rational integer b such that Q(J a ) = Q(d B ) . By Corollary 3.3.3, Q(JB)/Q is a Cogalois extension if and only if 6 # -1, -3. Since d/6 E @ , the result follows at once. 0 LEMMA 5.3.10. Let dl, dz E Q\Q2. Then Q (J$ , G ) / Q is a Cogalois extension if and only if -dl, -d2, -3d1, -3d2, -dld2, -3dld2 # @. PROOF.Set E = Q ( J&, &)/Q. If E I Q is a Cogalois extension, then its subextensions Q(J&)/Q Q( J&)/Q, and Q(J&&')/Q are also Cogalois by Proposition 3.2.2 (2), and so,
Q\@.
is a quadratic Cogalois extension if and only if d nE or 2(r d F d ) E @ and, either 2(r - d ) E @, and 2 ( ~ m - r ) 6 , (df-d-r) #@. (2) If [Qrgd: Q ] = 4 and d E Q then Qr,d/Q is a Cogalois extension if and only if -d, -3d, 2 ( - r & d F d ) , 6(-rf F d ) #
( 1Q
(3) If
[Qr.d
Q] = 4,
E Q(&)
\ Q,
and -d #
@,
then
Qr,d/Q is not a Cogalois extension.

(4) If [Qrgd: Q] = 4, d F d # Q(&), and either d m # Q( &) or dL- dr2 # @, then Qr,d/Q is not a Cogalois extension.
by Lemma 5.3.9. Conversely, assume that -dl, -d2, -3dl, -3d2, -d1d2, -3dld2 # @ . Then Q(d&)/Q Q(d&)/Q, and Q( JJ&)/Q are all Cogalois extensions by Lemma 5.3.9. If Q ( d & ) = Q(J&), then E = Q(J&, = Q ( d & ), and we are done. is So, we may assume that [ E : Q]= 4, and then, Q(J&, & ) / Q an Abelian extension with Galois group isomorphic t o Z2 x Z2, hence it has precisely three proper subextensions, namely Q(J& )/Q, J& )/Q, and Q(d=)/Q.
a)
PROOF. (1) By Proposition 5.3.3 and Remarks 5.3.4 (2), [ O r l d : Q] = 2 if and only if there exist c, k E (& such that r2 - d = c h n d r f c = k2/2. By Remarks 5.3.4 (2) and the proof of the implication (2) ==+ (3) in Proposition 5.3.3, we have
G (
If r - c = k2/2, then (r + c)/2 = (k2 + 4c)/4 Qr,d/Q is Cogalois in view of Lemma 5.3.9.
> 0, hence the extension
142
Chapter 5
143
Now, if r c = k2/2, then ( r - c)/2 = (k" 4c)/4, hence the extension is Cogalois if and only if 4c- k " @ and 3(4c- k2) $ @ , again by Lemma 5.3.9. Since 4c - k2 = 2(c - r ) = 2( J c d - r), we deduce that if Q , d /Q is a Cogalois extension, then 2 (dr"-d - r) , 6 ( d c d - r) $ @ . The conditions in the statement of (1) are also sufficient t o ensure that Qr,d/Q is a quadratic Cogalois extension.
(2) and (3): Assume that [Q,,d : Q ] = 4 and E Q(&). Then Qr,d/Q is a Galois extension by Proposition 5.3.7 (2). Two cases arise:
Thus, the extension Q T l d / Qwould be a Galois extension, which contradicts our assumption. Consequently, we must have Cog (QTld/Q)E Zq, hence there exists
a E Q such that
Then, again by Corollary 2.1.10, we deduce that { 1, fi, *2, *3 vector space basis of QT,d over Q. In particular, it follows that
} is a
Case (i): d we deduce that
m E Q.
Then, as in the proof of Proposition 5.3.6 (a)

: Q ] = 4, we deduce that fi # Q. Since Note that the Clearly, Q(JU) is a proper subfield of Q(S/Z) = cyclic group Cog ( Q , d / Q ) of order 4 has a unique proper subgroup, hence, by Theorem 3.2.3, it follows that Q J / Q has a unique proper intermediate subfield. Since Q(&) is also a proper subfield of QT,d,this implies = k J E for some k E Q. necessarily that Q( J Z ) = Q(&), hence Since QT,d = Q({Z) and f i = && ,there exist a , b, e, f E Q such that
where sary and sufficient, conditions for Qr,d / Q being Cogalois.
Q , , ~= ~Q ( J ( r + c ) , J 0 2 / , ) -~ c = J~2-d~. Now apply Lemma. 5.3.10 t o obtain the desired neces-
Case (ii): d m E Q ( J ~ ) \ Q Then . Gal (Q,,d/Q) E Z 4 by Proposition 5.3.7 (2). Assume that Qr,d/Q were a Cogalois extension. Then, by Corollary 5.2.3, we would have Then clearly exp(Cog (Qr,d/Q).) = 4, so i = E Q,,,l by Corollary 5.1.7. Thus. Q(i) and Q( 4 5 ) are quadratic subfields of Qr,d, and by Galois Theory, it follows that they are equal, so, in particular, we deduce that & = bi for some b E Q , ix., -d E @ , which contradicts our assumption. This proves that the extension Q r l d / Q is not Cogalois. (4) In case [Qr,d : 0 1 = 4 and dr"-d # ~ ( d d )then , Qr,d/Q is not a Galois extension by Proposition 5.3.7 (2). We shall adapt the idea of the proof of Proposition 3.2.6 (e) to this case. Assume that Qr,d/Q is a Cogalois extension. Since [ Q , d : Q ] = 4, then, by the definition of the concept of Cogalois extension, Cog (Q,,d/Q) is a group of order 4, hence it is isomorphic either t o Z 2 x Z 2 or t o Z4. If Cog (Qr,d/Q)
N Z2
c4
(12) 8:= J = = a + b f i + e f i + We can also write (12) as (13) 8-(a+efi)= Squaring ( U ) , we obtain *(b+
f&ifi.
f h ) .
e2 - ~ e ( a + e f i ) + ( a + e f i ) ~ = i & ( b + f & ) 2 .
Since 8' E Q( J 2 ) = Q(J Z ) , we deduce that we must have a+e& = 0, for otherwise, it would follow that 8 E Q(J Z ) = Q( Jd), which is impossible. Thus, (13) becomes which can be also written as
x Z2, then there exist
p, y E Q" such that
fi, is a vector space basis of Q , d over Q by CorolThen { 1, lary 2.1.10. In particular, it follows that
since
a, m }
fi = &/ k.
144
Chapter 5
6.3 Appiicatior~st o elerner~taryField Arithmetic (1)
145
From ( 1 5 ) we deduce that
Kow observe that the coefficient b2 in equation (17) is nonzero, for otherwise (17) would imply f = 0, and then (14) would become 8 = 0, which is impossible. So, the roots of (17) are
h , 2
bfr f b f d m b2
Q,
hence we must have d F d E Q , which contradicts our assumption. The roots of the quadratic equation (18) are Then, u coincides with its conjugate in the quadratic extension Q ( & ) / Q i.e.,
m E Q , which also contradicts our assumption. hence we must have d This completes the proof of the fact that Q r g d / Q is not a Cogalois extension. 0
This implies that
(bk+ f d q 2 r+&
hence
=&
(bk- f a ) ' r-Jd
'
COROLLARY 5.3.12. Let d E W, d 3 2 be a square-free integer, and let n E Z" be such that d r d # ~ ( d d Then ) . Q , J / Q 2s not a Cogalois extension. PROOF. By Corollary 5.3.5, [Q,,,d : Q ] = 4. According t o Theorem 5.3.11 ( 4 ) it is sufficient t o prove that d2 - dn" @ , or equivalently, that d2 - dn2 @ First, observe that " d dn2 # 0. Now, assume that d2 - dn2 E W 2 . Then, for any prime divisor p of d we must have p I d - n 2 since dL -dn2 = d(d - n " E E 2 and d is square-free, hence p I n? It follows that p I n, which implies that d I n since d is square-free. Thus n = ds for some s E Z*, hence
(16)
( b k + f&)2(r-&/;i)
= f( b k - f J E i ) 2 ( r + & ) .
v.
After easy calculations we obtain
if the sign in the right-hand side of (16) is
" + ", or
JZZZ=~J~TGEW,
if the sign in the right-hand side of ( 1 6 ) is " - ". Note that (14) implies that b+ f & #i 0, hence (16) can be also written as which implies that d m E N,i.e., 1 - ds2 = u 2 for some u E W .But d 2 2 by hypothesis, hence 1 = u " d s " 2, which is a contradiction. REMARK 5.3.13. The next example shows that we may have d2 - r 2 d E E Q\@ and r E If d E Q \ { - 1 ) is arbitrary, take r := 2 d / ( d I ) . Then d2 - r% = d2(d - ~ ) ~ / ( d + E @. If d = -1, take 0 r := 314. Then " d rr"d= ( 5 / 4 ) 2 .
@ for d
v.
Consequently, in case Qr,d/Q were a Cogalois extension, then necessarily we would have f(r2 - d ) E Q( & ) 2 , i.e., d r d E Q ( d d ) or d m E Q ( d d ) . Since d i # Q ( & ) by hypothesis, we deduce that the extension Qr,d / Q is not Cogalois whenever dd@ ~ ( d). d
Next, we investigate when is Q r , d / Q a radical extension. We will mainly discuss those which are subfields of R. Since any extension E I F with E a subfield of R is clearly pure and separable, by the GreitherHarrison Criterion it follows that E I F is radical if and only if it is Cogalois.
146
Chapter 5
6 . 3 Applications to elerrierrtary Field Arithrrretic (1)
147
Thus, the radical extensions of type Q r j d/Q, with r + the Cogalois ones.
> 0, are precisely
PROPOSITION 5.3.14. The following statements hold for r E Q and dQ\@. E Q then (),,d/Q is a radical Galois extension. (1) If (2) If \ / m e ~ ( d ; i ) \ ~d, > 0 , and r > 0, then QrId/Q is a nonradical Galois extension. (3) If d E N , d 2 2 is square-free, r E Z*, r + > 0 , and d c d # ~ ( d z )then , the extension Qr,d/Q is neither Galois nor radical.
There exist r E W and n l , . . . ,n,, a l , . . . , a, E N* such that K = Q ( n m , . . . ,n f i ) . The extension K / Q is radical. The extension K / Q is Kneser. The extension K / Q is Cogalois.
(4): The extension K / Q is clearly PROOF. (1) =j (4) and (2) separable, radical by hypothesis, and pure since K C R, hence it is Cogalois by the Greither-Harrison Criterion.
(4) (3) (2) are obvious. (1): Let {xl,. . . ,x,) be a set of representatives of the finite (4) -Xi (mod @ ), we may group Cog ( K l Q ) = T (K/Q)/Q* . Since xi assume that xi > 0 for all i, 1 i r . Then K = Q (xl , . . . ,x,), and for every i , 1 i r , there exists ni E N* such that xyi = ai E Q. Clearly, ai > 0 for all i. Then
+a
PROOF. If [Qr,d : Q] = 2, then Qrld is a quadratic field, hence Qr,d/Q is obviously a radical Galois extension. Next, we assume that [ Qr,d : Q] 5.3.4 (2). Two cases arise:
# 2. Then [ Q,,d : Q]
= 4 by Remark
< <
< <
Case 1: Qr,d/Q is a Galois extension. Then, by Proposition 5.3.7 (2) this is equivalent to d m E Q(&). There are two subcases: Subcase (i): d (a) we deduce that
K = Q("*,.
. . , ";/a,).
0
Of course, we may also assume that all ai E N*.
GE
Q. Then, as in the proof of Proposition 5.3.6
where c = d F d ,
SO
Qrtd/Q is a radical extension.
Subcase (ii): E Q(&) \ Q. Then, the extension Qrgd/Q is not Cogalois extension by Theorem 5.3.11 (3), so it is also not radical. Case 2: Qr,d/Q is not a Galois extension. By Proposition 5.3.7 (2), this is equivalent to \ / c d # Q ( d 8 ) . By Corollary 5.3.12, the extension Qr,d/Q is not Cogalois, hence it is not radical too. D
We end this section by showing that the abstract property of the ex&)/Q being not Cogalois, proved in Proposition 5.3.1 tension Q( J2 (e), can be equivalently expressed more attractively and elementarily as the impossibility to write d r f l as a finite sum of real numbers of type i "6, where r, n l , . . . ,n,, a l , . . . , a, E N*. The same problem will be +. discussed for any algebraic number a E W
COROLLARY 5.3.16. The following assertions are equivalent for a n algebraic number a E R t . (1) a can be written as a finite s u m of real numbers of type i "6, 1 < i < r , where r , n l , . . . ?n,, a1 ,... ,a, E N * . (2) The extension Q(a)/Q is radical. (3) The extension Q(a)/Q is Kneser. (4) The extension Q ( a ) / Q is Cogalois. PROOF. (1) ==+ (4): If cr is a number as in (I), then Q(a) is a subfield C R. By Proposition 3.2.2 (2), the extenof the field Q ( r L f i , . . . , sion Q ( " w , . . . , "-)/Q is Cogalois, hence, so also is its subextension
"6)
QW/Q
(2) follow from Proposition 5.3.15. (4) u (3) (4) ==+ (1): -4pply Proposition 5.3.15 t o the real algebraic number field K = Q ( a ) t o deduce that Q(a) = ~("'fi,.
.. ,
"'a),
PROPOSITION 5.3.15. The following statements are equivalent for a real algebraic number field K .
for some s, m l , . . . ,m,, bl, . . . , b, E N*. Then a is a sum with coefficients hence it has the desired form. in Q of products of powers of '"6, COROIJARY 5.3.17. The following assertions hold.
148
Chapter 5
5.4 1Sxercises t o Chapter 5
149
be written as a finite sum of real ;umbers of type i "i/ii;, 1 i r, where r, 7x1,. . . ,n,, a l , . . . , a, E N*. (2) Let d 2 2 be a square-free integer, and let r E Z* be such that r > -& and J T d $ Q(&). Then cannot be written as a finite sum of real numbers of type f "6, 1 i r , with T, n1,. . . ,n,, a l , . . . ,a, E N* .
< <
and call the polynomial r = ( X - yl ) ( X - & ) ( X - y3) the cubic resolvent of f . Prove the following statements. (a) If f = X 4 ax" bX" CX d, then (b) If (c) If f E K [ X ] is an irreducible polynomial, and is any of the four roots XI, 22, 23, x4 E fl o f f , then the extension F ( c ) / F has no proper subextension if and only if the cubic resolvent of f is irreducible in F[X].
< <
+ + r = x3- bx2+ (ac - 4d)X - (a% - 4bd + 2). f = x4+ m X + n , then r = X 3 - 4 n X - m 2 .
PROOF. Apply Corollary 5.3.16, Proposition 3.2.6 (e), Remark 5.3.2, and Proposition 5.3.14 (3). 0

1. Prove that the following statements hold for a Galois G-Cogalois
extension EIF, with HI I F * , . . . ,H,/F* the Sylow subgroups of GIF* (a) F ( H i ) / F is a Galois extension for every i, 1 i r. (b) Gal ( E I F ) is a nilpotent group.
< <
2. Let E/F be a G-Cogalois extension, and let El IF, E2/Fbe subextensions of E / F . Establish the possible implications between the statements below. (a) El n E2 = F . (b) The fields El and E2 are linearly disjoint over F . (c) Kne(E1E 2 / F ) r Kne(E1/ F ) x Kne(&/F). 3. caicui,e
7. (Kappe and Warren [75]). Let F be any field, let f = X 4 + a x 3 + bx2+ c X + d E F [ X ] be a separable irreducible quartic polynomial, and let r be the cubic resolvent of f . Let E be a splitting field of f , let K be a splitting field of r , and let D be the discriminant of f . Denote by M4 the alternating group of degree 4 and by 9 4 the dihedral group of order 8. Prove that the following statements hold. (a) Gal ( E I F ) r 6 4 u r is irreducible in F [ X ] and D $ F 2 . (b) Gal ( E I F ) LX M4 u r is irreducible in F [ X ] and D E F 2 . (c) Gal ( E I F ) E Z 2 x Z2 r splits into linear factors over F. (d) Gal ( E I F ) LX Z 4 U r has exactly one root y E F , and the polynomial (X" yX + d)(X2 + a x + (b - y)) splits over K. (e) Gal ( E I F ) 2 D 4 u r has exactly one root y E F and the
polynomial ( X L y X + d)(X2 t a x over K.
+ ( b - y))
does not split
8. Deduce the result of Exercise 5 from Exercises 6 and 7. 9. Let d < 0 be a square-free integer, and let n E Z . Show that Qn,d/Q is a Galois extension if and only if n2 - d E N*2.
10. Show that the following statements hold for a given square-free positive integer d 3 2. (a) If d is even, then there is no n E Z such that Qn,d/Q is a Galois extension with Galois group isomorphic t o Z 2 x Z2. (b) If d is odd, then there exist only finitely many n E Z such that Q,,,d/Q is a Galois extension with Galois group isomorphic to Z2 x Z2. (c) There exists a t least an n E Z such that Qn,d/Q is a quartic cyclic extension if and only if the Pell-Fermat equation x " dy2 = -1 has a solution (x, y) E Z* xZ*. In this case n = dy, and in fact there exist infinitely many such n.
[Q(
J J -
: Q]
4. Calculate [ Q (
: Q]
, where r t
a.
<
5. Let n E Z , and let J be a not specified complex root of the polynomial X 4 - n X - 1. Prove that the field Q(J) is a quartic field having only two subfields if and only if n E Z \ { -4, 0, 4 ).
6. Let F be any field, let s2 be an algebraically closed overfield of F , and let f E F [ X ] be a quartic polynomial having distinct roots 21, 22, 23, 2 4 E 0. Denote
91 = x1x2 +x3x4, y2 =Xlx3 +x2x4,
y 3
= 21x4 +x2x3,
150
Chapter 6
5.5 l%ibliographicalcorr~rr~ents to Chapter 5
151
11. Let r E Q and d E Q \ @ . Prove the following statements. (a) If c := d m E (& and 2(r - c) E @ , then
19. If r E W , then show that
(b) If c := d
mE Q
(I + i) e Z4
Cog(Qr,d/Q) = {4, (c) If c : =
and 2(r + c) E @, then
r radicals
(ifi.(l+ifi))
d m }
if 2(c - r ) E @ Z Z 6 if 6 ( c - T ) E @ otherwise. -3d
2 Z2
4 -
q, 2 ( r k c ) , 2 ( - T I C ) , 6 ( - T ~ c ) ,-d,
6 @ , then
12. Let r E Q and d E Q \ @ . Prove the following statements. (a) If J c d E Q( 4 2 ) \ Q, [QT,d : Q ] = 4, and the extension Qr,d/Q is Cogdois, then Cog (Qr,d/Q) = (1 + i)
20. Let r E W , r 2 3, and set q = 2'. Prove that the following statements hold. (a) Q(&) = Q(i,q), where q = ( , [;' = 2cos(7r/2'-'). (b) Gal(Q(cq)/Q) = G1 @ G2, where GI is the cyclic subgroup of order 2T-2 of Gal(Q(G )/Q) generated by the automorphism o E G a l ( ( ) / ) defined by o l ( G ) = [ : , and G2 is the cyclic subgroup of order 2 of Gal(Q([,)/Q) generated by the = ([I. auto~norphism0 2 E Gal(Q(&)/Q) defined by oz (6) (c) G a l ( Q ( c o s ( ~ / 2 ~ - l ) ) / Q =)G I .
21. Let n E N. Prove that cos(7r/2") can be written as a sum of real numbers of type f "\./a,, 1 6 i r, where r, nl , . . . ,n,, a , , . . . , a, E N* , if and only if n E (0, 1, 2). 22. Prove (a) (b) (c) that the following statements are equivalent for an n E Q(cos ( ~ / 2 ~ ) ) is lQ a Cogalois extension. Q(cos(7r/2")) /Q is a Kneser extension. Q(COS ( 7 ~ / 2 ~ ~ ) is )/Q a radical extension. (4 n E (0, 1, 2).
= Z4.
N.
(b) If [ Q r , d : Q] = 4, and the extension Qr,d/Q is not radical, then
be such that dr"-d E Q, [Qr,d : Q] = 4, 13. Let r E Q, d E Q \ and such that the extension Qr,d/Q is not Cogalois. Calculate cog(Q,d/Q). 14. Calculate Cog(Q ( 4 & , a)/()), where d l , d2 E Q
23. For which n E
is Q(sin(71-/2"))/Q a Cogalois extension?
24. Show that Cog (Q(cos(?i/2"))/Q) = {T, fi}for any n E N, n
> 2.
\ @.
15. For which d l , d2, ds E Q \ Q is Q ( d & , extension? 16. Calculate Cog(() ( 4 6 ,
a ) / Q a Cogalois
a &)/Q),
where dl, dz, ds E Q \
@.
Section 5.1. The results of this section are taken frorn Albu [8], Albu and Kicolae [19], and Albu and Tena [25]. Section 5.2. Theorem 5.2.2, due to Albu a d Nicolae [19], generalizes Corollary 5.2.3 due to Barrera-.-R4ora,Rxedowski-Calderh, and Villa,Salvador 1301.
17. Investigate whether or not the conditions in Theorem 5.3.11 (3)-(4) are also necessary for Qr,d/Q being a non Cogalois extension. 18. Investigate whether or not the condition " r removed in Proposition 5.3.14.
+& !>
0 " can be
152
Chapter 5
Section 5.3. The results of this section are taken from Albu [8], [7], and Albu and Panaitopol [24].
CHAPTER 6
RADICAL EXTENSIONS AND CROSSED HOMOMORPHISMS

In this chapter we investigate finite Galois extensions which are radical, Kneser, or G-Cogalois, in terms of crossed homomorphisms. The results of this chapter are based on the description, provided in Section 6.1, of the Cogalois group Cog(E/F) of any finite Galois extension EIF by means of crossed homomorphisms of the Galois group G a l ( E / F ) with coefficients in the group p ( E ) of all roots of unity in E . This description, which is actually a reformulation of the Hilbert's Theorem 90 in terms of Cogalois groups, states that there exists a canonical group isomorphism
A consequence of this result, of the uniqueness of the Kneser group of a G-Cogalois extension, as well as of the n-Purity Criterion, is the description of the Kneser group of any finite Galois G-Cogalois extension EIF with the aid of crossed homomorphisms of G a l ( E / F ) with coefficients in the group p,(E) of all n-th roots of unity in E , where n is the exponent of the Kneser group G I F * of E I F . Another nice application of the Cogalois-like reformulation of Hilbert's Theorem 90 is the finiteness of the Cogalois group of any finite extension of algebraic number fields. In Section 6.2 we characterize via crossed homomorphisms finite Galois extensions which are radical, Kneser, or G-Cogalois, and provide sufficient conditions under which the property of a Galois extension EIF being radical, Kneser, or G-Cogalois is preserved when one changes the base field F.
154
Chapter 6
6.1 Galois extensions and crossed homomorphisms
155
6.1. Galois extensions and crossed homomorphisms In this section we describe first the Cogalois group of a finite Galois extension EIF as the group of crossed homomorphisms of the Galois group Gal ( E I F ) with coefficients in the group p ( E ) of all roots of unity contained in E . A consequence of this fact is the description of the Kneser group Kne ( E I F ) of any Galois G-Cogalois extension by means of crossed homomorphisms. As an application of this result, one deduces very easily that the Cogalois group of any finite extension of algebraic number fields is a finite group. R.ecal1 first only those basic facts on Galois Cohomology which will be used in the sequel. Let EIF be an arbitrary extension with Galois group I ' ,and let M E* be such that a ( M ) M for every a E I'. A crossed homomorphism (or an 1-cocycle) of I ' with coefficients in &I is a map f : I' + M satisfying the condition
For every positive integer n 2 1 we shall define the following subgroups of T ( E lF ) and Cog(E/F), respectively. T n ( E / F ) = { x E E* I xn E F*), C o g n ( E / F ) = Tn ( E I F ) I F * . For a n arbitrary extension E/F we consider the following map f : Gal ( E I F ) x Cog ( E I F ) + p ( E ) , f(a,G) = fa(u) = a(a).a-', a E Gal(E/F), a E T ( E / F ) . Note that f is well-defined. Clearly, for every integer n 2 1, the restriction of f t o Cogn(E/F) induces a map For every fixed a E Gal ( E I F ) , the partial map f (a, -) is clearly multiplicative on Cog ( E lF ) , and for every fixed 3 E Cog(E/ F ) , the partial map f (-, G) is precisely the 1-coboundary fa E Z1 (Gal ( E I F ) , p(E)), so f and frl induce the group morphisms
<
f ( 0 4 = f ( 4 . a ( f (4), for every a , T E I ' . The set of all crossed homomorphisms of I ' with coefficients in M is an Abelian group, which will be denoted by Z ' (T, M ) . For every a E M we shall denote by f, the 1-coboundary f a : I '+ M , defined as f a ( a ) = a ( a ) . a - l , a E r. The set B1(I', M) = { fa I a E M ) is a subgroup of Z 1 ( r ,M ) . The q~iotient group Z1 (I', M ) / B 1 (I', M) is called the first cohomology group of I ' with coefficients in M , and is denoted by H1( I ' , M). Note that for any group G, for any G-module A, and for any n E N one can define the Inore general concept of n-th cohomology group H n ( G ,A) of G with coefficients in A (see e.g., Cassels and Frohlich [46, Chapter IV] or Karpilovsky [76, p. 3691). The famous Hilbert 's Theorem 90 asserts that H 1 (Gal(E/F), E * ) = 1 . for any finite Galois extension EIF (see e.g., Cassels and Frijhlich [46, Proposition 2.2, Chapter V] or Karpilovsky [76, Theorem 9.2, Chapter 61).
Recall that for any extension EIF we use throughout this monograph the following notation. p ( E ) = { x E E* I x 7 ' = 1 for some n E W), T ( E / F ) = { x E E* I x n E F* for some n E W ), Cog(E/F) = T ( E / F ) / F * , 2 = the coset x F * E E*/F* of any x E E .
respectively.
LEMMA 6.1.1. For any finite Galois extension EIF and for any n E
I T , the morphism
$71
: Cogn ( E I F ) + Z1(Gal ( E I F )
1171
( E ))
defined above is a group isomorphism.
PROOF. Denote by I ' the group Gal ( E I F ) , and let a E T,(E/F). We have
6 E Ker($,) M f a ( a ) = 1, Va E
and consequently,
$lr,l
r w a(a) = a ,
Va E
e a E Fix ( r ) = F,
is a monomorphism. We are going to prove that $,, is surjective. Let h E Z 1 ( r , p n ( E ) ) . E * ) , by Hilbert's Theorem 90 there exists a E E* Since clearly h E Z1 (I', such that h = fa. Hence a ( a ) . a - ' E p,,(E), so ( ~ ( a ) . a - ' )= ~ 1 for every a E I ' . We deduce that ~ ( a =~ a ~ n for ) every a E r , and consequently
156
Chapter 6
6.1 Galois extensions and crossed homornorphisrr~s
157
an E F* since EIF is a Galois extension. Thus, cu E T n ( E / F ) . Hence 0 h = @,(2), with d E Cogn(E/F). This shows that &, is surjective.
is a lattice isomorphism, which induces a canonical lattice isomorphism

Subgroups(Cog ( E I F ) )
E
Subgroups ( 2 ' (I?, p ( E ) ) ) .
THEOREM 6.1.2. For any finite Galois extension E / F , the assignment d I+ f, establishes a group isomorphism
Cog(E/F)
E 2'
(Gal(E/F), p ( E ) ) .
For evey cyclic subgroup C of Z1 (I?, p ( E ) ) there exists a E T ( E I F ) such that p ( F *( a ) ) = ( f, ) = C.Moreover, H I F * ES p ( H ) for eve? H 0 with F* 6 H T ( E / F ) .
<
PROOF. Clearly, for every n 2 1, the restriction of the morphism

$ : Cog(E/F) 4 2' (Gal(E/F), p ( E ) )
t o Cogn(E/F) is &. On the other hand, since I? = Gal(E/F) is finite, Z1(I?, p ( E ) ) is the union of all Z1 (I?, p n ( E ) ) , n 2 I , and since every t j ~ , is surjective by Lemma 6.1.1, we deduce that II, is also surjective. 0 REMARK 6.1.3. We will see in Section 15.1 that if E/F is an infinite Galois extension, then the group Cog(E/F) is isomorphic t o the group of all continuous crossed homomorphisms of the compact topological group G a l ( E / F ) (endowed with the Krull topology) with coefficients in the discrete group p ( E ) . 0 COROLLARY 6.1.4. Let E / F be a finite Galois extension with p ( E ) finite. T h e n Cog(E/F) is a finite group. I n particular, for any extension K / L of algebraic number fields, which is not necessarily Galois, the group Cog(K/L) is finite.
LEMMA 6.1.6. Let E/F be a G-Cogalois extension, let x E R* be such that x7'& E F for some m E W , and let n = exp ( G I F * ) . Suppose that one of the following two conditions is satisfied. (1) P, C PTl ( i n particular, this holds if m 1 n ). (2) pm ( E ) C F (in particular, this holds if , ( E F). Then, we have F ( x ) C E (jx E G.
PROOF. Suppose that F ( x ) C E. Set K = F ( x ) and k = ord(2). Then clearly k = exp ( F *(x)/F*) and k 1 m . Let p E P k . Then p 1 m, and so, p 1 n if condition ( I ) is satisfied. We deduce that hence p E Pm7 p p ( K ) C p p ( E ) F, since E/F is n-pure by Theorem 4.3.2. If condition (2) is satisfied, then we have /L,(K) p p ( E ) C p m ( E ) E F . Hence, in both cases, K / F is F*(x)-Cogalois, again by Theorem 4.3.2. But, the extension K / F is also G n K*-Cogalois by Proposition 4.3.5, so F * ( x ) = G n K * by Corollary 4.4.2. Thus x E G, as desired. The other implication is 0 obvious.
1
II
I ~
PROOF.Since Gal(E/F) and p ( E ) are finite groups, it is obvious that the group 2' (Gal(E/F), p ( E ) ) is finite, hence Cog(E/F) is also finite by Theorem 6.1.2. Since K / Q and L / Q are both finite extensions, it follows that the extension K / L is a finite separable extension. Consider the normal closure k / of ~the extension K/L, which is a finite Galois extension (see 1.2.7). Then, c o g ( i / L ) is a finite group since p ( N ) is a finite group for any algebraic number field N . Now, observe that Cog(K/L) is a subgroup of the finite group c o g ( k / ~ )hence , it is also finite. 0
THEOREM 6.1.7. Let EIF be a Galois G-Cogalois extension with n = exp(G/F*). Then, there exists a canonical group isomorphism
PROOF. Denote by r the Galois group of E/F. Let a E T,(E/F). By Lemma 6.1.6, we deduce that a E G, so G = T n ( E / F ) . Thus, and so, the desired isomorphism follows a t once from Lemma 6.1.1. For any Galois G-Cogalois extension E/F with n = exp(G/F*), the map f : Gal(E/F) x Cog(E/F) 4 p(E), considered at the beginning of this section yields by restriction the map
158
Chapter 6
6 . 2 Radical extensioris via crossed hornomorphisnis
159
For any A ( Gal ( E I F ) and any W
< Kne ( E I F ) let denote
6.2. Radical extensions via crossed homomorphisms

In this section we investigate via crossed homomorphisms when a finite Galois extension is radical, Kneser, or G-Cogalois. We also give sufficient conditions under which the property of a Galois extension E / F being radical, Kneser, or G-Cogalois is preserved if one changes the base field F . Let EIF be a finite Galois extension with Galois group I?. By Theorem 6.1.2, there exists a canonical isomorphism Cog ( E I F ) Z' (I?, p ( E ) ) , hence the canonical map
wT = { o E Gal ( E I F ) I g(o, c) = 1, Vc E W }.
PROPOSITION 6.1.8. For any Galois G-Cogalois extension E I F , the assignments (-) define mutually inverse antGisomorphisms between the lattices Subgroups (Gal ( E I F ) ) and Subgroups - (Kne(E/F)).
PROOF. For simplicity, denote by I' the group Gal ( E I F ) , by C the 1 the lattice of all subgroups of G I F * = lattice of all subgroups of r , by 3 Kne ( E I F ) , and by E the lattice of all intermediate fields of the extension EIF. Since EIF is a G-Cogalois extension, the maps E -+ 31, K
H
f : Gal ( E I F ) x Cog ( E I F ) + p ( E ) ,
f ( 0 , s ) = .((.).a-',
considered in Section 6.1 yields, by replacing Cog ( E I F ) with its isomorphic copy Z' (r, p ( E ) ) , precisely the evaluation map For any A ( r , U ( Z1 ( r , p(E)), and
( K n G ) / F * , and 31 -+ E, H I F * ct F ( H ) ,
x E Z1(I',p ( E ) ) denote
are isomorphisms of lattices, inverse to one another. On the other hand, since EIF is a Galois extension, the maps C -+ E, A ct Fix(A), and E
-+
C, K u G a l ( E / K ) ,
{ O E1 ~ (o,~)=l).
are anti-isomorphisms of lattices, inverse to one another. If 4 6 r and W = H I F * 6 G I F * , where F* H ( G, then it is easily proved that
<
One verifies easily that A' Z1(I',p(E)), U'I ' , and = (x)'. but Kate that in the previous section we have also used the notation (-)I, with a different meaning. The next result characterixes radical subextensions of a given Galois extension EIF by means of subgroups of Z1(Gal ( E I F ) , p(E)).
<
<
and
={o
r1
o(8) =0, V
~ H E} = G a l ( E / F ( H ) ) .
It follows that each of the composed maps 31--+E-+C and C-+E-+31
THEOREM 6.2.1. Let EIF be a finite Galois extension with Galois group I?, and let K be a n intermediate field of E I F . T h e n K l F is a radical extension (resp. a simple radical extension) if and only if there ex' ists U Z1 (I?,p ( E ) ) (resp. x E Z' ( r , p ( E ) ) ) such that Gal ( E I K ) = U (resp. Gal ( E / K ) = X I ) .
<
of the canonical bijections considered above gives rise t o the maps (-)T between C and 31, which finishes the proof. 0
PROOF. If K I F is a radical extension, then there exists a group G, not necessarily unique, such that F* G T ( E / F ) and K = F ( G ) . If we set := {fa I ( . E G } 6 zl(L p(E)), then we have
< <
160
Chapter 6
6.2 Radical extensions via crossed homornorphisrrls
161
= { a E I'
o(x) = x, Vx E F ( G ) ) = Gal ( E / F ( G ) ) = Gal ( E I K ) .
Conversely, suppose that there exists an U Z1(I?, p ( E ) ) such that Gal ( E I K ) = UL. Let G = { a E E*I f a E U ) ( T ( E / F ) be the group of radicals associated with U by Corollary 6.1.5. We have just seen that U = Gal ( E / F ( G ) ) , hence Gal (E/K)=Gal (E/F(G)). By the Fundamental Theorem of Galois Theory we obtain K = Fix (Gal ( E I K ) ) = Fix (Gal ( E / F ( G ) ) ) = F ( G ) , which shows that K I F is a G-radical extension. The case of simple radical extensions now easily follows from Corollary 6.1.5. 0 REMARK 6.2.2. Denote by Radical ( E I F ) the set of all subextensions K / F of E / F which are radical. With the notation and hypotheses of Theorem 6.2.1, the map Subgroups (2' (I', p ( E ) ) ) -+ Radical ( E lF ) , U Z1(r, @)I, is surjective in view of Corollary 6.1.5. In general, this map is not injective (see Exercise 4). This may happen because distinct subgroups G and G' of T ( E / F ) containing F * , may define the same field F (G) = F ( G i ) . The result below provides characterizations, in terms of crossed homomorphisms, of Kneser and G-Cogalois subextensions of finite Galois extensions. COROLLARY 6.2.3. Let EIF be a finite Galois extension with Galois group I ' , let K / F be a finite G-radical subextension of E/F with G I F * a finite group, and denote U = { f, I a E G } Z 1 ( T , p ( E ) ) . Then (1) The extension K I F is G-Kneser if and only if (I': UL) = JUI. (2) The extension K / F is G-Cogalois if and only if it is G-Kneser and the map
<
(2) By Theorem 4.3.2, the extension K I F is G-Cogalois if and only if it is G-Kneser, and the map
is canonically isomorphic to the lattice { V I V ( U }. On the other hand, the canonical map Intermediate ( K I F ) + {A 1 U L
< G } --+Intermediate ( K I F ) , H H F ( H ) , is a lattice isomorphism. Since U I G I F * , the lattice { H I F* < H < G }

{ H IF* ( H
< A ( I' }, L I+ Gal(E/L),
is an anti-isomorphism of lattices by Galois Theory. Summing up, we deduce that the G-Kneser extension K I F is GCogalois if and only if the canonical map obtained by composing the three canonical maps considered above is bijective, or equivalently, an anti-isomorphism of lattices. Observe that the last part of the proof of Theorem 6.2.1 shows that b is precisely the map 0 V I+ v'. Next, we give sufficient conditions under which the property of a Galois extension EIF being radical, Kneser, or G-Cogalois is preserved when one changes the base field F . Recall that R denotes an algebraically closed field containing the given field F as a subfield. Any overfield L of F which will be considered in the sequel is supposed t o be a subfield of 0. Let E/F be a finite Galois extension with Galois group I ' , let L I F be any extension with L n E = F , and consider their compositum E L . -4ccording t o the Galois Theory (see 1.2.9), the restriction map Gal ( E L I L ) -& Gal ( E I F ) , o
r-, a
u I+ F({ a E E*I fa E U ))IF,
<
<
l~,
is an isomorphism of groups. Consequently, by the Fundamental Theorem of Galois Theory we deduce that the maps
E:
is bijective, or equivalently, an anti-isomorphism of lattices.

PROOF. (1) The extension K I F is G-Kneser if and only if [ K : F ] = IG/F*I. If we set A = Gal(E/K), then [ K : F] = (I' : A ) by Galois Theory, and G I F * s U, so (G/F*I = IUI, by Corollary 6.1.5. On the other hand, A = U' according t o Theorem 6.2.1. Summing up, we deduce that [ K : F ] =IG/F*I # (I?: A) = lUl # (I':u')
= lUl.
Subextensions ( E I F ) -+ Subextensions (ELIL), K I F
LKIL,
and
A: Subextensions (ELIL) ---t Subextensions ( E I F ) , K I I L

are isomorphisms of lattices, inverse t o one another
H (K1 n E ) / F ,
We have denoted in Remark 6.2.2 by Radical(E/F) the set of all subextensions K I F of EIF which are radical. For every K I F E Radical ( E I F ) there exists a group G, not necessarily unique, with F* ( G T(E/F)
<
162
Chapter 6
6.2 Radical extensions via crossed hornornorphisn~s
163
and K = F(G). If we set G1 = GL*, then clearly L K = L(G1) and L* 6 GI 6 T(EL/L). It follows that E ( K / F ) E Radical (ELIL), and consequently, the restriction of e t o radical extensions gives rise t o the injective map p : Radical ( E I F ) -+ which is not necessarily bijective. Radical (ELIL),
Fix(Ga1 (ELIL)) = L . We deduce that cul = a . y for some y E L* G I . L*. (Gl nE*) 1 .y-l E Gl nE * , which proves the inclusion G1 Thus a = a
THEOREM 6.2.4. Let EIF be a finite Galois extension with Galois group I?, and let L I F be an arbitrary extension such that E n L = F . If p ( E L ) = p ( E ) , then the following assertions hold. (1) GL* n E* = G for every G with F* 6 G T ( E / F ) . (2) G1 = (GI n E*)L* for every GI with L* 6 GI 6 T(EL/L). (3) The map
(3) Though apparently the definition of p as given in the statement of the theorem seems t o depend on the group G, which is not necessarily uniquely associated with a given radical subextension F (G)/ F of E I F , we have seen above that p is a well-defined injective map. To prove that p is a surjective map, let K1/L E Radical (ELIL). Then K1 = L(Gl) for some G I with L* 6 G1 T ( E L / L ) , hence, if we set G = G1 n E * , then F ( G ) E Radical ( E I F ) , and
<
p ( F ( G ) / F ) = L ( F ( G ) ) / L = L ( F ( G I n E * ) ) / L = L((GI n E*)L*)/L, L((G1n E * ) L * )= L(G1) = L(GL*) = K1, which shows that p is surjective, hence bijective, and its inverse map p-' is that described in the statement of the theorem. Observe that
p-' : Radical (ELIL) + Radical ( E I F ) ,
<
p : Radical ( E I F ) -+ Radical (ELIL),
can be also defined as
is bijective, and the map

Radical (ELIL) -+ Radical ( E I F ) ,
(K1I L ) = (Kl n E ) / F .
is its inverse.
PROOF. (1) follows a t once from the equality E n L = F .

(2) Denote
REMARKS 6.2.5. (1) The isomorphism v defined in the proof of Theorem 6.2.4 induces the isomorphism of lattices
rl = Gal (ELIL). We have seen that the map
(*)
{U
1 u 6 w"p(E)))
U
+ {Ul I Ul 6 . W ' , P ( E L ) )
H U1
1 7
= v(U).
is an isomorphism of groups. Since p(EL) = p ( E ) , this isomorphism induces the group isomorphism v(h)(al1 = h(al IE), h E z1 ( r , P(E)), 01 E r l . Let G1 with L* 6 GI 6 T(EL/L). The inclusion (G1 9E * ) L * G1 is obvious. Now let a1 E G I . Then f,, E Z1(T1,p(EL)), hence there exists a E T ( E / F ) such that f,, = v(f,) since v is an isomorphism and Z1(T,p(E)) = B1 (I',T(E/F)) by Theorern 6.1.2. Thus, fa1 ( a l ) = fa (01 [E) for all a1 E TI. It follows that o l ( a l ) . a;' = 01 ( a ) a-l for all 01 E r l , i.e., al (a1 a - ' ) = a 1 . a-' for a11 a1 E r l , hence al - a-' E
By Corollary 6.1.5, there exist lattice isomorphisms
{ u p 6 .W"I.@)))
+ {GIF* 6 G ,<T(EIF)h UI-+G={~EE*I~,EI/),
Row, using (*), we obtain an isomorphism of lattices u:{GIF* 6G6T(E/F))+{G11L* < G I <T(EL/L)),
164
Chapter 6
6.2 Radical exter~sions via crossed horr~orr~orphisn~s
165
We claim that v(G) = GL* for every G with F* G T(E/F). Indeed, if G1 = v(G), then it is easily seen by the above considerations that for an a1 E (EL)* one has
<
<
[ F ( G ) : F] = (I' : UL) = (I'l : u ( U ) l ) = (I'i : U,I) = [L(G1) : L ] . (2) By Corollary 6.2.3 (I), the extension F ( G ) / F is G-Kneser if and only if (I' : UL) = IU 1, and similarly, the extension L(G1)/L is G1-Kneser ) : = IUl I. But Ul = v(U), so IU1I = IU I. To conclude, if and only if (I'l : u apply (1). (3) By Corollary 6.2.3 (2), F ( G ) / F is G-Cogalois if and only if it is G-Kneser and the map is bijective. Using (2) and the canonical isomorphism G1 E G, these conditions are equivalent to the conditions that L(Gl )/L is a GI-Kneser extension and the map is bijective. Again by Corollary 6.2.3 (2), these last conditions are equivalent to the fact that the extension L(G1) / L is GI-Cogalois. 0 COROLLARY 6.2.7. Let EIF be a finite Galois extension, and let L / F be an arbitrary extension such that L n E = F . Let K be an intermediate field of the extension E/F such that K L / L is a G1-radical extension, with G1/L* a finite group of exponent n. If Cn E E , then K / F is G-radical and G / F * E G1/L*, where G = Gl nE*.
Hence GI = v(G) = GL*. From Theorem 6.2.4 (2) we deduce that the inverse v-' of v can be described explicitly by v-l(G1) = Gl
n E * , L* < G1 < T(EL/L).
(2) Another argument for the surjectivity of p is the following one. If K I / L E Radical (ELIL), then there exists Ul Z1(Gal (ELIL), p ( E L ) ) such that Gal (EL/K1) = U? by Theorem 6.2.1. Let
<
be the isomorphism considered in the proof of Theorem 6.2.4 (2), and denote U = u-' (Ul ). It is easily verified that Gal ( E / ( K l n E ) ) = UL. Again by Theorem 6.2.1, we deduce that (Kl nE ) / F E Radical ( E I F ) , and by Galois Theory we have p((K1 n E ) / F ) = K1/L. 0 COROLLARY 6.2.6. Let E/F be a finite Galois extension, and let L / F be an arbitrary extension such that L nE = F and p ( E L ) = p ( E ) Let G be a group such that F* G T ( E / F ) , and denote G I = GL*. Then (1) G / F * Z G l / L * and [ F ( G ) : F] = [ L ( G l ) : L]. (2) The extension F ( G ) / F is G-Kneser if and only if the extension L(G1)/L is G1 -Kneser. (3) The extension F ( G ) / F is G-Cogalois if and only if the extension L(G1 ) / L is GI -Cogalois.
< <
PROOF. We have already noticed that the restriction map to E gives rise to an isomorphism
I ' , = Gal (ELIL) l i Gal ( E I F ) = I?.

Let Ul = { f,, I a l E G1 ) be the subgroup of Z1 (I'l, p ( E L ) ) which corresponds by Theorem 6.2.1 t o the radical subextension L(G1)/L of the Galois extension EL/L. Since a? E L* for every a1 E G1, it follows that fa, (ul) E pn(fl) for every a 1 E I ? ' , and consequently, Ul Z1 ( r l , ( C , , ) ) . By Theorem 6.2.1, we have U t = Gal(EL/L(Gl)). As in the proof of Theorem 6.2.4, the above considered isomorphism
PROOF. (1) We have G1/L* = (GL*)/L* r G/(G n L*) = G / F * since G n L* = F. Let U := {fa'[ a E G } be the subgroup of Z1(r, p(E)) which corresponds via Theorem 6.2.1 t o the radical subextension F ( G ) / F of the Galois extension E / F , and preserve the notation from the proof of Theorem 6.2.4. If we denote Ul = v(U), then by Theorem 6.2.1, we have
<
yields the group isomorphism

+
166
Chapter 6
Denote U = v-I (UI). Since Cn E E, it follows that U Z1(I', p(E)). If we denote G = { a E E*I fa E U ), then, as in Remark 6.2.5 (I), we deduce that G = GI n E * , G I = GL*, and so, Finally, by the proof of Corollary 6.2.6 (1) we have G / F * Z G I /L*. COROLLARY 6.2.8. With the notation and hypotheses from Corollary 6.2.7, the extension K/F is G-Kneser (resp. G-Cogalois) if and only if K L I L is G I -Kneser (resp. G1-Cogalois). PROOF. Adapt the proof of Corollary 6.2.6. The details are left t o the 0 reader.
<
4. Let F = Q,E = Q(C3), and and I Radical(E/F) 1 = 2 .
= Gal(E/F). Show that
5. Show that the result of Lemma 6.1.6 may fail if neither condition (1) nor condition (2) is satisfied. (Hint: Take the Q ( 0 ) - C o g a l o i s extension Q(C3)/Q and the element C3.)
6. Let E/F be a G-Cogalois extension with n = exp (GIF*), and let x E T ( E / F ) with m = ord(2). Prove that x E G if and only if
Pm C pn*
7. Show that Lemma 6.1.6 holds even if m = 2.
P ,
= 0 , i.e., if m = 1 or
6.3. Exercises to Chapter 6 1. (Dummit [55]). Let EIF be a finite Galois extension, let r = Gal(E/F), and let A a I'. Show that the canonical sequence of Abelian groups is exact, where P(E)* = { C E p ( E ) 1 u(C) = C, V u E 4 ). Show that the right-hand map is not generally surjective, even if I' is a cyclic group. 2. Let E/F be a finite Galois extension, and let K be any intermediate field of E / F . Show that the diagram below Cog(E/F) Cog(E/K)
8. Let K be an algebraic nurnber field, let r, n l , . . . ,n, E W with n l , . . . , n, 2 2, and set 12 = n l . . . n,. Prove that there exist % Z/niZ for a l , . . . , a, E N* such that aai E K*", K*"(ai )/K*7L every i E (1,. . . , r ) , and
r
K*"(al,
,a,)/K*"
@ (K*"(ai)/K*"),
2=
that is, for any k,,. . . , k, E a?' . . . . . aFr E K*"
N one has
+=+ ni I ki for all i E {1, . . . ,r).
Z1( G a W I F ) , P ( E ) ) Z1 G a W I K ) , P ( E ) ) is commutative, where all the arrows are canonical morphisms.

3. Let E/F be a finite Galois Cogalois extension with Galois group I'. Prove that the canonical morphism
is surjective for every A
(Hint: Let bK be the discriminant of K , and let p l , . . . ,p, be distinct prime numbers which do not divide b K . Then p, , . . . ,p, nln; are unramified primes in K, and take ai := pi , i = 1 , . . . ,r . )
< I'. (Hint: Use Exercise 2.)
9. Let K / F be a separable extension of degree n 2 2 with gcd(n, e ( F ) ) = 1, and set L = F(Cn) and E = K(Cn). Suppose that K nL = F , E I L is an Abelian extension, and E/F is a Galois extension. Prove the following statements. (a) E I L is an Abelian extension with exp(Gal(E/L) In, so E I L is a classical n-Kummer extension. (b) There exist r, n l , . . . ,n, E N* with n l , . . . ,n, 2 2 and a l , . .. , a r E L* such that, if we set n = nl . . . . . n,, then [ L ( m ) : L ] = ni = o r d ( m ) for every i, 1 i r, E = L( ... , E*''nL* = L * " ( a l , . . . ,a, ), , . . . ,a, are independent modulo L*", that is,
m, m),
a:'
< <
. . . . a;. E L*" w ni I ki for all i = 1 , . . . ,r,
Chapter 6
6 . 3 Exercises t o Chapter 6
169
and where for every i # j in (1,. . . ,r). r, one has (c) For every a E Gal(L/F) and every i, 1 i a ( a i ) E L*"(al, . . . ,ar ), hence there exist uniquely determined elements L and 0 b s < n ~ ,. . , 0 b;, < n, such that
(Hint: Apply Exercise 9.) 13. (Barrera-Mora and Vdlez [32]). Let El /F and E2/F be finite separable subextensions of an extension E I F , with El n Ez = F . Suppose that the extensions E2/F and El E2/E2 are both Galois. Prove that E1E2/Fis a Galois extension.
< <
<
<
(Hint: Use Hasse [69, Satz 152, p. 2231 .) 10. Let K / F be a separable extension of degree n 2 2 with gcd(n, e ( F ) ) = 1, and suppose that K n F(C,,) = F . Prove the following statements. (a) If K (Cn)/F(C,,) is an Abelian extension, then K (c,,)/ F is a Galois extension (b) There exist s E N* and a l , . . . ,a, E F* such that K = F( . . . , i/a,) if and only if K (Cn)/F(cn) is an Abelian extension and, with the notation of Exercise 9, the following condition is satisfied: (t) b$ = dij for all a E Gal(F(C,)/F) and i, j E (1, . . . ,r ) . (Hint: Use Exercises 8 and 9, and Theorem 6.2.1.)
14. (Barrera-Mora and Vdlez [32]). Let K / F be a separable extension of degree n 2 2 with gcd(n,e(F)) = 1, and suppose that K n F(Cn) = F and K(cn)/F(Cn) is a cyclic extension. Prove the following statements. (a) K(Cn)/F is a Galois extension. Moreover, K(Cn)/F is an Abelian extension if and only if K I F is a cyclic extension. 1(mod n) (b) K / F is a simple radical extension if and only if b , for all a E Gal(F(Cn)/F), where the integers b, were defined in Exercise 12. (Hint: Apply Exercises 12 and 13, and Theorem 6.2.1) 15. (Barrera-Mora and Ve'lez [32]). Let E/F be a separable extension of degree a prime number p different from Char(F), and suppose that the following conditions are satisfied. (a) F(Cp)/F is a quadratic extension, (b) The normal closure of EIF is E(Cp)/F, (c) Gal(E(Cp)/F) is isomorphic t o the dihedral group apof order 2p. Prove that EIF is a simple radical extension. (Hint: Use Exercise 14.) 16. (Barrera-Mora and Vdlez [32]). Prove that for any field F of characteristic # 3, and for any separable cubic extension EIF such that E(C3)/F is the normal closure of E I F , there exists a E F* such (Hint: Use Exercise 15.) that E = F(*). 17. (Barrera-Mora and Vdlez [32]). This example shows that condition (c) in Exercise 15 cannot be omitted. Let u = (1 &)/2, a = $hi7 F = Q(C5), and E = F ( a ) . Prove the following statements. (a) u E F, M i n ( a , F ) = x5- u , and Min(a,Q) = XI0 - X5 - 1. (b) E / Q is a Galois extension with Gal(E/F) = ( a , T), where o ( & ) = (z, r ( & ) = 5 j , a4 = r5 = lP;, and a 0 7 = r 30 0 . (c) If K is the fixed field of the 2-Sylow group of Gal(E/Q), then [ K : Q] = 5 and E / Q is the normal closure of K / Q . (d) K / Q is not a simple radical extension.
m,
11. (Barrera-Mora and Vdlez [32]). Let EIF be a separable extension of degree n ) 2 with gcd(n, e(F)) = 1, and let E/F be the normal closure of E I F . Suppose that there exists a finite extension L / F such that E ( & ) n L = F and L E = L ( = ) for some a E L*. (Hint: Apply Theorem 6.2.1.) Prove that E = F ( fi). 12. (Barrera-Mora and Ve'lez [32]). Let K / F be a separable extension of degree n 2 2 with gcd(n, e ( F ) ) = 1, and set L = F(C,) and E = K(C,,). Suppose that K n L = F , E / L is a cyclic extension, and E/F is a Ga.lois extension. Prove the following statements. (a) There exists a E L* such that E = L(*) and Gal(E/L) = ( T ) , where r ( @ ) = Cn*. (b) For every a E Gal(L/F) there exist y, E L and b, E N : . abs . such that gcd(b,, n) = 1 and a ( a ) = y
170
Chapter 6
6.4 13ibliographical corr~n~er~ts to Chapter 6
171
18. (Barrera-Mora [29]). An extension E/F is said to be a repeated radical extension if there exists a tower of fields
IFi is a simple radical extension for all i = 0 , . . . ,n - 1. so that Fi+1 Show that a subextension of a repeated radical extension is not necessarily a repeated radical extension. (Hint: Any cubic cyclic extension K/Q is not a repeated radical extension, but it is a subextension of the repeated radical extension K(C3)/Q.)
19. (Barrera-Mora [29]). Show that there exists a repeated radical extension E/F having an intermediate field K such that the extension K / F is not a repeated radical extension. Furthermore, the extension E/F can be chosen t o be Galois and E \ F contains no roots of unity. 20. (Barrera-Mora [29]). Let E / F be a repeated radical Galois extension such that E \ F contains no p t h root of unity for any prime number p. Prove that F contains a primitive p t h root of unity for every prime divisor p of [ E : F ] . 21. (Barrera-Mora [29]). Let E/F be a repeated radical extension so that E \ F contains no p-th root of unity for any prime number p . Prove that there exists a tower of fields
(b) There exist intermediate fields E l , . . . ,E, of E/F such that Ei/F is a r d i c a l tower and [ Ei : F] = for all i = 1,. . . ,T. Prove that every subextension of E/F is a radical tower.
pri
24. (Barrera-Mora [29]). Prove that under the assumptions of Exercise 23, the fields E l , . . . ,E, are unique. 6.4. Bibliographical comments to Chapter 6
') such that [ Fi+,: Fi1 is a prime number pi and Fi+l = Fi( & for some ai E Fi7i = 0,. . . , n - 1.
22. (Barrera-Mora [29]). A radical tower is a repeated radical extension E/F such that there exists a tower of fields with Fi+l Fi(n&), ai E Fi, and [Fi+l: Fi] = ni for all i = O , ... , n - 1. Show that in a radical tower, the ni's can always be taken t o be prime. 23. (Barrera-Mora [29]). Let E / F be a separable extension of degree n ) 2, and let n = p:l . . . . plr, with p l , . . . ,p, mutually distinct prime numbers and kl ,. . . , k, E N* . Assume that the following conditions are satisfied. (a) Cpi $ E \ F for all i = 1 , . . . , r .
Section 6.1. Theorem 6.1.2 was established in the particular case of finite Galois extensions of algebraic number fields by Dummit [55], and for arbitrary Galois extensions which are not necessarily finite by Barrera-Mora, bedowski-Calderh, and Villa-Salvador [30] (see Theorem 15.1.2). Our approach, based on Lemma 6.1.1 is a bit different, and can be also easily adjusted for infinite Galois extensions. Corollary 6.1.4 is essentially due to Greither and Harrison [63], who proved it for finite extensions of algebraic number fields by using a different approach. The idea to use Theorem 6.1.2 for the proof of the generalization of Greither and Harrison's result to arbitrary finite Galois extensions E / F with p ( E ) finite is due to Barrera-Mora, Rzedowski-Calderbn, and VillaSalvador [30]. Theorem 6.1.7 is due t o Albu and Nicoiae [22], and is based on an older version of Lemma 6.1.6, proved by Albu, Kicolae, and Tena [23]. Lemma 6.1.6, as stated in this section is due t o Albu [7]. Corollary 6.1.5 and Proposition 6.1.8 were established by Albu and Kicolae [22]. Particular cases of Proposition 6.1.8, for E/F a Galois Cogalois extension or a so-called "neat presentation" were proved by Greither and Harrison [63] using a quite sophisticated technique including the Lyndon-Hochschild spectral sequence. Section 6.2. All the results of this section are due to Albu and Kicolae [22]. Theorems 6.2.1 and 6.2.4 generalize to arbitrary finite radical extensions similar results of Barrera-h4ora and VCllez [32] established for simple radical extensions.
CHAPTER. 7
EXAMPLES OF G-COGALOIS EXTENSIONS

The fundamental prbperty of Cogalois extensions says that any such extension E/ F is an extension with T ( E lF)/F*-Cogalois correspondence (Theorem 3.2.3 (1));in other words, an extension with T(E/F)-Cogalois extension. We show in this chapter that this is an immediate consequence of the n-Purity Criterion we proved in Section 4.3. We deduce the essential part of the finite Kummer Theory from the fact that any finite classical Kummer extension is G-Cogalois, which in turn, is an immediate consequence of the n-Purity Criterion. Moreover, this criterion allows us to provide other large classes of G-Cogalois extensions which generalize or are closely related to classical finite Kurnmer extensions: finite generalized Kummer extensions, finite Kummer extensions with few roots of unity, and finite quasi-Kummer extensions.
7.1. Classical Kummer extensions In this section we present first various characterizations of classical Kurnmer extensions which are not necessarily finite. Then, we show that finite classical Kummer extensions are G-Cogalois. As an immediate consequence of this fact we deduce the whole finite classical Kummer Theory. Recall that R is a fixed algebraically closed field containing the fixed base field F as a subfield; any considered overfield of F is supposed t o be a subfield of R. For any nonempty subset A of F* and any n E W we will denote by the subset of T ( f l / F ) defined by
n:= { x E Rlz" E A ) .
7.1 Classical Kurnrr~er extensions
175
In particular, if A is a singleton {a}, then is precisely the set of all roots (in a ) of the polynomial X7' - a E F[X]. We shall use throughout this monograph the notation to designate a root, which in general is More precisely, for not specified, of this polynomial. Thus, @ E any choice of the root @i, we have
m.
= F ( @). In particular, if 6 E F , then F ( However, in certain cases, for instance when F is a subfield of the field R of all real numbers and a > 0, then @i will always mean the unique positive root in R of the polynomial X n - a.
m)
of n , and so [ K : F] In. But, it is known that if CTl E F , then for any finite cyclic extension K / F of degree a divisor of n , there exists an x E K * such that x7' E F and K = F ( x ) (see 1.2.11), in other words, K I F is a simple radical extension F ( @ ) / F generated by an n-th radical & of some a E F * . Set B = { x 1 x E E * , xn E F ) and denote by B' the set of all elements in B that generate those cyclic subextensions K / F of E/F whose compositum is the given extension E/F. Since B' B, we have E = C_ E, hence E = F ( B ) . If we set A = B n , then clearly F ( B ) F(B1) E =~ ( m ) .
DEFINITION 7.1.1. A classical n-Kummer extension, where n E W ,is an Abelian extension E/F such that gcd(n, e ( F ) ) = 1, p,(fl) E F and G a l ( E / F ) is a group of exponent a divisor of n . A classical Kummer extension, or just a Kummer extension is any extension which is a classical n-Kurnmer extension for a certain integer n 3 1. If E/F is a classical Kummer extensiont we also say that E is a classical Kummer extension of F. 0 The next result gives a more precise form of classical Kumrner extensions which are not necessarily finite.
(1): Obviously, condition (2) implies that E is the composi(2) tum of the family of fields ( F ( & ) ) a E A . AS is well-known, each extension F ( *)IF is a cyclic extension of degree a divisor of n , hence EIF is an Abelian extension. Let a E Gal(E/F). Then, for every a E A, the restriction of an t o F ( @ ) is the identity map, which implies that un = l E . This shows that the group Gal(E/F) has exponent a divisor of n. (2)
(3) is obvious.
THEOREM 7.1.2. The following assertions are equivalent for an extension E/F and a natural number n 2 1. (1) E/F is a classical n-Kummer extension. ( 2 ) gcd(n,e(F)) = 1, pn(R) C F , and E = F ( 0 ) for some 0 # A g F*. (3) gcd(n, e ( F ) ) = 1, pn(R) C F , and E = F ( B ) for some 0 # B C E* with Bn C F .
PROOF. (1) ==+(2): The extension E/F is the compositum of all its finite subextensions E 1 / F . Observe that the finite Abelian group Gal(E1/F) is an internal direct sum of a finite family (Hi),<i<7, of cyclic subgroups, where each Hi is the Galois group of a cyclic subextension K i / F of E / F . We deduce that each such finite extension E1/Fis a compositum of cyclic extensions (see 1.2.9), hence E/F is the compositum of all its cyclic subextensions. Let K / F be an arbitrary finite cyclic subextension of E / F . Since G a l ( E / F ) is by hypothesis an Abelian group of exponent a divisor of n, it follows that a'' = l Kfor all a E Gal(K/F), hence IGal(K/F) I is a divisor
The form of finite classical Kurnmer extensions is an immediate consequence of Theorem 7.1.2. COROLLARY 7.1.3. The following assertions are equivalent for a finite extension E/F and a natural number n 2 1. (1) E/F is a classical n-Kummer extension. (2) gcd(n,e(F)) = 1, pn(R) C F , and E = F ( m , . . . ,*) . . ,a,) F * . some r E W and {al,.
for
0
We are now going t o show that the whole finite Kummer Theory can be very easily derived from Cogalois Theory. A similar result also holds for infinite Kummer extensions and will be presented in Section 13.1.
LEMMA 7.1.4. Let E/F be a separable G-radical extension with G I F * finite, and let n E W be such that G7' F * . If the extension E/F is n-pure, then EIF is G-Cogalois.
PROOF.The finite group G / F * has a finite exponent, say m, and m I n by Lemma 1.4.6. If p E P , , , then p 1 n, and p,(E) F by hypothesis. Consequently, the extension E / F is also na-pure, hence it is G-Cogalois by Theorem 4.3.2. 0
176
Chapter 7
7.1 Classical Kumrrier extensioris
177
LEMMA 7.1.5. The following statements hold for any extension E/F such that p n ( E ) F for a given n E N*, and any group G such that F* < G < E * . (1) The map G / F * -+ Gn/F*", x F * u xnF*", is a group isomorphism. (2) The maps H u H7' and M u i/M n G establish lattice isomorphisms, inverse t o one another, between the lattices
(3) If H is any subgroup of F * 7 L ( a l , ... ,a,) containing F * n , then any is a vector space basis set of representatiues of the group of F ( q ) over F , and [ F ( D ) : F] = IH/F*"I. I n particular, one has
m/~*
[ E : F] = IF*(%,.
F*(i/aT,.
.. , s ) / F * I = I F * n ( a l , .. . ~ , ) l F * ~ l .
Hom(Gal(E/F), p n ( F ) ) .
( 4 ) There exists a canonical group isomorphism
{ H I F * < H < G ) and { M ( F * " < M < G n ) .

(3) If pn(R)
. . ,= ) I F *
c F , then
c G for
every
# X 2 Gn .
PROOF. (1) The surjective group morphism G duces clearly a surjective group morphism
h
-+ Gn, x u xn, in-
PROOF. (1) Set A = { a l , . . . ,a,), B = and G = F * ( B ) . Since pn(R) F it follows that the extension EIF is n-pure. Clearly Gn E F , hence the extension EIF is G-Cogalois by Lemma 7.1.4. But G = F* (B) = F*( .. . , and the result follows.
c m,
s),
n)
(2) By (1) and Theorem 4.3.2, the maps
For simplicity, denote i? = x F * and i? = x F *for ~ any x E E * . We are going t o show that cp is also injective. Let x E G be such that 4 9 ) = i. Then, ? i = P xn E F*" j xn = yn for some y E F* + ( x ~ - ' )= ~ 1 j XY-' E pTl(E) xy-' E F* I$ x E F* j 2 = 7.
A
H u F ( H ) and K u K n ( ~ * ( ) i ; /
are lattice isomorphisms, inverse to one another, between the lattice of all subgroups H of F*( containing F*and the lattice of all subextensions K / F of E / F . If K / F is any subextension of E / F , then it is an Abelian extension and its Galois group is a quotient group of Gal(E/F), hence a group having exponent a divisor of n. By definition, K / F is a classical n-Kummer extension. Now, apply Lemma 7.1.5 for G = F * ( and observe that G n = F*"( A ).
(2) R.ecal1 that for any nonempty subset A of F , we denoted ; / ; ; I = { x 1 x E R, x7' E A ) , where f l is an algebraically closed field containing E as a subfield. Using again the fact that p n ( E ) C F , one shows easily that
a)
;/ZT;;~G = H and (;/ii?n~)"

for every F* H
=M
< < G and every F*"< M < Gn, which proves (2). (3) Clearly, X c Gn implies that E m. But = G.
Indeed, the inclusion G is clear. For the opposite inclusion, let x E Then x7' = gn for some g E G, hence (xg-l)n = 1, i .e., xg-' E pn ( 0 ) C F. Consequently, x E G since F* G. 0
(3) The extension E/F is F * ( ; / ; ; I )-Cogdois by ( I ) , hence the extension FF /) ( ? f/; is i/l?-~neser by Corollary 2.1.10. Vse again Lemma 7.1.5 and Corollary 2.1.10 to deduce the desired statements. (4) We have Kne(E/F) = G / F * = F * ( . . . , *)IF* by (1). On the other hand, by Theorem 6.1.7, there exists a canonical group isomorphism Kne(E/F) 2 2' (Gal(E/F), p,(E)), where m = exp(G/F*). Since E/F is a classical n-Kummer extension we have p,(R) = p7'(E) = pT1(F) S F , and since m 1 n we also have / L ~ ( EC ) p n ( E ) C F. But Gal(E/F)) r Kne(E/F) by (1) and Theorem 5.2.2, hence 2' (Gal(E/F), P,,,(E)) = Hom(Gal(E/F), p,, ( E l )
= Hom(Gal(E/F), pn ( F ) ) and the desired result follows.
m.
m,
THEOREM 7.1.6. Let EIF be a finite classical n - K u m m e r extension, with E = F ( ; / i i ; ,. . . ,fi), n , r E K,and { a l , . . . ,a,) C F * . Then, the following statements hold. (1) E/F is a n F* ( ,. . . , fi)-Cogalois extension. (2) The maps H u F( m ) and K u K n n ( F * " ( a l , . . . ,a,)) establish isomorphisms of lattices, inverse to one another, between the ( a1 , . . . ,a,) containing F*nand lattice of all subgroups H of F*" the lattice of all intermediate fields K of E I F . Moreover, any subextension K / F of EIF is a classical n - K u m m e r extension.
178
Chapter 7
7.2 Generalized Kunirner exterlsior~s
179
7.2. Generalized Kummer extensions In this section we present another class of G-Cogalois extensions, which is larger than the class of Kummer extensions, namely the class of generalized Kummer extensions. This new class also includes the class of Kummer extension with few roots of unity which will be discussed in the next section. We show that a theory of finite generalized Kummer extensions, which is very similar to that of finite classical Kummer extensions, can be developed using the properties of G-Cogalois extensions; since, in general, they are not Galois extensions, no other approach (e.g., via Galois Theory, as in the case of classical Kummer extensions) is applicable. The case of infinite generalized Kummer extensions will be examined in Section 13.2. DEFINITION 7.2.1. We say that a finite extension E/F is a generalized n-Kummer extension, where n E W , if gcd(n, e ( F ) ) = 1, p,,(E) 5 F , andthereexist r N ' , a l , . . . , a , F* such that E = F ( m ,... ,-).
(3) If H is any subgroup of F*7L( a1 , . . . ,a,) containing F*", then any n G ) / F * is a vector space set of representatives of the group basis of F ( m n G ) over F , and [ F ( f / i ? n G ) : F] = IH/F*"J. In particular, one has
(n
PROOF. Proceed as in the proof of Theorem 7.1.6, by applying Lemma 7.1.5 and Lemma 7.2.2. To see that any subextension K I F of EIF is still a generalized nKummer extension, observe that, by (I), K = F ( U ) for some subgroup F * , and so, any element Gn U of G containing F * . Clearly Un
<
of the finite group U/F* is a finite product kl , . . . ,k, E N. Since clearly p n ( K ) 5 p,,(E) K/F is a generalized n-Kummer extension.
k1
. . . . fi , where
one concludes that
kr.
5 F,
A generalized Kummer extension is an extension which is a generalized n-Kummer extension for some integer n 2 1. 0
Clearly, any classical n-Kummer extension is a generalized n-Kummer extension. By Corollary 5.1.4, any generalized Kummer extension is a separable extension, but not necessarily a Galois extension. LEMMA 7.2.2. Let EIF be a finite generalized n-Kummer extension, r E W*, and a l ,... ,a, E F * . Then E/F is with E = F ( m ,... ,*), ( fi, . . . ,fi) -Cogalois extension. an F* PROOF. Observe that the extension EIF is G-radical, where G = F* ( *, . . . ,f i ) , and Gn F; it is also n-pure, since for every p E Pn one has pp(E) E p n ( E ) S F . It follows that E/F is a G-Cogalois extension by Lemma 7.1.4, as desired. 0
The next result establishes when a generalixed Kummer extension is a classical Kurnmer extension.
PROPOSITION 7.2.4. A finite generalized Kummer extension is a classical Kummer extension if and only if it is a Galois extension.
PROOF.Let E/F be a finite generalized n-Kummer extension, with E = F ( m ,... ,*), r E N*,a , , . . . ,a, E F * , and p,(E) E F . Denote andlet m=exp(G/F*). G = F * ( % , ... ,*) If E/F is a Galois extension, then gcd(m,e(F)) = I and E E by Corollary 5.1.7. But m I n , hence
c,,
Thus, E/F is a classical m-Kummer extension. Conversely, any classical Kummer extension is by definition a Galois extension. 0 REMARK 7.2.5. Let EIF be an extension such that there exist r E W , n1, . . . ,n r E W* and a l , . . . ,a, E F* with E = F ( "*,. .. , Let G = F*( .. ,? y & ) and n = lcm ( n l , . . . ,n,). We claim that if gcd(e(F), n) = 1 and p n ( E ) C F , then EIF is a generalized n-Kurnmer extension and a G-Cogalois extension. Indeed, for each i, 1 i T , there exists qi E W* such that n = niqi. Hence we can ' = 6 and then E = F ( a , . . . , with Ci = a:' for write ' all i, 1 i 6 r . By Lemma 7.2.2, E/ F is an F* ( . . . ,i/c,)-Cogalois 0 extension, so, it is also a G-Cogdois extension.
THEOREM 7.2.3. Let F be a field, and let r, n E W , a l , . . . ,a, E F*. Denote G = F * ( m , . . . ,-) and E = F(C/li;, . . . ,f i ) . I f gcd (n, e ( F ) ) = 1 and pn ( E ) C F , then the following statements hold. The generalized n-Kummer extension El F is G- Cogalois. The maps H ct ~ ( m n and ~ K ) ct K n n ( F * " ( a l , ... ,a,)) establish isomorphisms of lattices, inverse to one another, between the lattice of all subgroups H of F * n (a1 , . . . ,a r ) containing F*n and the lattice of all intermediate fields K of E I F . Moreover, any subextension K / F of E/F is a generalized n-Kummer extension.
nm,.
< <
"6).
I,
<
m,
m,
s),
180
Chapter 7
EXAMPLES 7.2.6. ( 1 ) A generalized Kummer extension is not necessarily a classical Kurnmer extension (see Exercise 2). ( 2 ) Observe that any generalized n-Kummer extension E I F is n-pure since for any p E P, one has p p ( E )E pn(E) C_ F . However, the converse 0 is not true according t o Exercise 3.
basis of F ( m n G ) over F , and [ F ( mn G ) : F ] = IH/F*"I. I n particular, one has
Recall that whenever a E R; and n E W , then @ will always mean the unique positive root in R of the polynomial Xrl - a.
7.3. Kummer extensions with few roots of unity

In this section we deal with finite Kummer extensions with few roots of unity, which are very particular cases of generalized Kummer extensions discussed in the previous section.
7.3.1. A finite extension E I F is said t o be a n n-Kummer DEFINITION extension with few roots of unity, where n E N*, if gcd(n, e ( F ) ) = 1, p n ( E ) C (-1, I ) , and there exist r E W , and a l , . . . ,a, E F* such that E = F ( m , . . . ,*). A Kummer extension with few roots of unity is a n extension which is a n n - K u m m e r extensions with few roots of unity for some positive integer 12 2 1. 0
The next result establishing the basic properties of Kummer extensions with few roots of unity is an obvious consequence of Theorem 7.2.3. We state it for the reader's convenience.
COROI~LARY 7.3.3. Let F be a subfield of R, let r, n E W , and let a l , . . . ,a, E F* be positive real numbers. Denote G = F * ( m , .. . ,fi) and E = F ( . . . ,S). Then, the following statements hold. ( 1 ) The extension E I F is G-Cogalois. ( 2 ) The maps H u ~ ( m n and ~ K u ) K n n ( F * " ( a l ,... ,a,)) establish isomorphisms of lattices, inverse to one another, between the lattice of all subgroups H of F * " ( a l , . . . ,a,) containing F*" and the lattice of all intermediate fields K of E I F . then any ( 3 ) If H is any subgroup of F*"( a1 ,. . . ,a,) containing F*7L, n G ) / F * is a vector space set of representatives of the group basis of F ( m n G ) over F , and [ F ( mn G ) : F ] = IH/F*rlI. I n particular, one has
m,
(m
PROOF. The result follows irnrnediately from Theorem 7.3.2 since any subfield K of R is obviously a field with few roots of unity, i.e., p ( K ) C
THEOREM 7.3.2. Let F be a field, and let r , n E N*, a ] ,. . . ,a, E F*. Denote G = F * ( . .. , and E = F ( m , . . . , If gcd (n, e ( F ) ) = 1 and p n ( E ) C (-1, I ) , then the following statements hold. ( 1 ) The n - K u m m e r extension with few roots of unity E I F is GCogalois. ( 2 ) The maps H u F ( m n G ) and K ct K n n ( F * " ( a l , .. . ,a,)) establish isomorphisms of lattices, inverse t o one another, between the lattice of all subgroups H of F*n( a l ,. . . ,a,) containing F*" and the lattice of all intermediate fields K of E I F . Moreover, any subextension K I F of E I F is an n - K u m m e r extension with few roots of unity. ( 3 ) If H is any subgroup of F * n ( a l , .. . ,a,) containing F*", then any n G ) / F * is a vector space set of representatives of the group
{ - I , 1).
m, s)
s).
REMARK 7.3.4. The extension Q (J-3, fi)/Q (J-3)is an example of a classical Kummer extension which is not a Kummer extension with few roots of unity (see Exercise 6). 0
7.4. Quasi-Kummer extensions

The aim of this section is t o present another class of G-Cogalois extensions which is close t o the class of classical Kummer extensions.
(m
7.4.1. W e say that a finite extension E I F is a quasiDEFINITION Kummer extension if there exist r E W , n l , . . . ,n, W , and a l , . . . ,a, E F* such that E = F ( I & . . ~ . ,, lAC/a,), gcd(n, e ( F ) )= 1, and p p ( R ) C F for any p E P,,, where n = lcm (nl ,. . . ,n,). 0
182
Chapter 7
Observe that the condition pp(fl) 2 F is equivalent t o the condition
Cp
E F , and that any finite classical Kummer extension is a quasi-Kummer
course, we may suppose that ai E Z* for all i = 1 , . . . , r , and we may ignore those ai for which ni = 1. (3) ==+ (4) follows from Corollary 7.1.3. (4) ===+ (2) and (2) ===+ (1) are obvious.
extension. However, the class of quasi-Kummer extensions is strictly larger than the class of classical Kummer extensions (see Exercise 9).
THEOREM 7.4.2. A quasi-Kummer extension F ( " m , . . . , "&)IF as in the Definition 7.4.1 is F* ( " m , . . . , ";/a,)-Cogalois.
n = lcm(n1,. . . ,n,), and PROOF. Let G = F*("&, . . . , "&), m = exp(G/F*). Then m I n, hence for any p E P7, we have p 1 n , and so, p p ( E ) pp(fl) F . This shows that E/F is m-pure. On the other hand, since gcd (m, e ( F ) ) = 1, the F*("*,. . . , "&)-radical extension E/F is separable by Corollary 5.1.4. Now apply Theorem 4.3.2, t o conclude that the extension E/F is G-Cogalois. 0
COROI,I~ARY 7.4.5. The cyclotomic extension Q((,)/Q is G-Cogalois for some group G if and only if n E {1,2,3,4,6,8?12,241. PROOF. Assume that the extension Q([,,)/Q is G-Cogalois for some group G. As is well-known, this extension is an Abelian extension with
THEOREM 7.4.3. Any finite Galois G-Cogalois extension is a quasiKummer extension.

PROOF. Let EIF be a finite Galois G-Cogalois extension, and consider a set {bl,. . . , b,) of representatives of the finite group G I F * . If n = exp ( G / F * ) then by = ai E F for all i , 1 6 i 6 r , hence Since the G-radical extension EIF is Galois, it follows that gcd (n, e ( F ) ) = 1 and C,, E E by Corollary 5.1.7, so p,,(fl) p,(E). By Theorem 4.3.2, the G-Cogalois extension E / F is n-pure, hence pp(E) C F for all p E P,. Thus, pp(fl) p p ( E ) F for all p E P,. This shows that the extension E/F is quasi-Kummer. 0
where, as usually, U(Z,,) denotes the group of units of the ring Z7, of integers modulo n. By Theorem 5.2.2, we have G/Q" Gal(Q(Cn )/Q) 7
hence G / p U(Z,,). On the other hand, by Corollary 7.4.4, Q(C,,)/Q has to be a classical 2-Kummer extension, which is H-Galois with respect t o a group H with e x p ( H / Q ) 2. It follows that G = H by Corollary 4.4.2. Consequently, we must have e x p ( G / V ) 2, i.e., every element of li(Z,,) must have order 6 2. . . . . - p$ with Of course, we may assume that n 2 2. Let n = 2k k, s, kl, . . . ,k, E N and pl < . . . < p, positive odd prime numbers. According t o the Chinese Reminder Theorem, there exists a ring isomorphism
<
<
C OROILARY 7.4.4. The following assertions are equivalent for an algebraic number field E . (1) E / Q is a Galois G-Cogalois extension for some group G. (2) E/Q is an Abelian G-Cogalois extension for some group G. (3) There exist finitely many nonzero rational integers a1, . . . ,a, such that E = Q ( d G , . . . ,&). (4) E / Q is a classical 2-Kummer extension.
i
which yields a group isomorphism
It is known that U(Z2) and
{O), li'(Zd)
Z2, U(Z2k)
Z2 X z2k-2, k 2 3,
PROOF. (1) 3 (3): By Theorem 7.4.3, E / Q is a quasi-Kummer extension, hence there exist r, n l , . . . ,n, E N*, and a l , . . . ,a, E such that E = Q("*, . . . , ";/a,) and pp(C) C Q for every p E P,, where n = lcm ( n l , . . . ,n,). But C7, E Q if and only if m = 1 or m = 2. Consequently, we must have n 2, hence ni 2 for all i = 1 , . . . r. Of
U(Zps) Z,.-I(,_,), p odd prime, s E W* . All these facts imply that there exists a group isomorphism G / ~ r ~ ( ~ 2 * ) x ~ p : 1 - l (xp... , I XZp:D-'(p )
),
"
<
<
where U(Z2k) is one of the groups {O), Z2, or Z2 x Z2k-2 when k 2 3.
184
Chapter 7
7.5 Cogalois extensions
185
But the group U(Z2k) should not contain any element of order > 2, hence necessarily k - 2 6 1, i.e., k E {O,1, 2,3). Assume that s 3 1 and let i, I 6 i s. Since the group Z k i - I P; (pi-1) should not contain any element of order > 2, it follows that ki 1 and pi - 1 = 2, hence pi = 3. But p1 < . . . < p,, hence necessarily s = 1 and pl = 3. Consequently,
<
THEOREM 7.5.1. Any finite Cogalois extension EIF is T ( E / F ) Cogalois.

PROOF. By the Greither-Harrison Criterion (Theorem 3.1.7), the extension E/F is separable and pure, so, a fortiori, n-pure, where n = exp ( T ( E / F ) / F * ) . By Theorem 4.3.2 we deduce that E/F is T ( E / F ) Cogalois. 0 We end this section by discussing an analogue of the Inverse Galois
<
Conversely, for any of the listed values of n we are going t o express Q(cn)/Q as a quasi-Kummer extension and indica.te the group G for which the extension is G-Cogalois. For n = 1 and n = 2 we have Q(C1) = Q(c2) = Q, and then G = Q . For n = 3 we have Q(&) = Q(J-3) and G = Q ( G ) . For n = 4 we have Q(c4) = Q(i) and G = p (i). For n = 6 we have Q(c6) = Q(J-3) and G = O+ (G ). For n = 8 we have Q(C8) = Q(42 i fi) = Q(4 2 , i) , and then
Theory Problem for Cogalois extensions.
P~orosr~ro 7.5.2. x Given any finite Abelian group A, there exists an algebraic number field E such that E / Q is a Cogalois extension and Cog(E/Q) E A.
PROOF. According t o the structure theorem of finitely generated Abelian groups, we have
r
k=1
G = Q( h , i ) . For n = 12 we have Q(C12)= Q(J3+ i) = Q(J3,i), and then
for some r, n l , . . . ,n, E N*. Choose r different positive prime integers p l , . . . ,p,, take and let,
G = Q( h , i ) .
For n = 24 we have Q(Gn) = Q ( C ~ S12 + isin
7r
-) 12
7r
= Q((JG
+ JZ) + i(&
JZ))
G = Q ( nfi,. . . , rLG) and n = lcm(n1,. . . ,n,).

Then, E / Q is a classical n-Kummer extension with few roots of unity, so, by Remark 7.2.5, it is G-Cogalois. The extension E / Q is also Cogalois, and Cog(E/Q) = G / Q by Corollary 4.4.3. Since p l , . . . ,p, are different . . . ,( is an independent prime numbers, we deduce that ( "i/ii;), family of subgroups of the group Q ( . . , T L @ ) / Q (use e.g., Lemma 9.2.4), hence
and then G = Q
(a, &,i). The proof is complete.

7.5. Cogalois extensions
= Q( J 2 , &,i).
- "m)
n
We gave in Section 3.2 a straightforward proof of one of the basic properties of Cogalois extensions, namely that saying that any such extension EIF is an extension with T(E/F)/F*-Cogalois correspondence. We show in this section that, with the aid of the n-Purity Criterion (Theorem 4.3.2), this property can be deduced immediately by placing the Cogalois extensions in the general framework of G-Cogalois extensions. We also investigate when a given finite Abelian group can be realized a s the Cogalois group of a Cogalois or a Galois Cogalois extension.
But
& n
Q( y & . .
. , ni/j9/Q' E! ( ~lfi) x . .. x ( y / q .
?
is a cyclic group of order ni for every i = 1.. . . ,r , so we have Cog(E/Q) Z,, x . . . x Z , , 2 , . A,
and we are done.
PROPOSITION 7.5.3. Let r be an arbitrary finite group. Then, there ! r if and only zf exists a Galois Cogalois extension E / Q with Gal(E/Q) " I ' (Z2)' for some r E N .
186
Chapter 7
7.6 Nxercises to Chapter 7
187
PROOF. If E / Q is a Galois Cogalois extension, then it is a Galois

T(E/Q)-Cogalois extension, and so, by Corollary 7.4.4, we deduce that
E/Q is a classical 2-Kummer extension. Consequently, Gal(E/Q) r (Z2)' for some r E N.

Conversely, suppose that I ' E (Z2)7'for some r E N. Of course we may assume that r 2 1. Choose r different positive prime numbers pl , . . . ,p,, and let E = Q(JpT,. . . Then E / Q is a Galois Cogalois extension 0 such that Gal(E/Q) E ( 2 4 ' .
5. Show that any finite classical ZKummer extension E/F is a 2Kummer extension with few roots of unity, but it may happen that P , ~ ( E ) F for some n E W , n # 2.
6. Let F = Q(G) and E =
,a).
~ ( fShow i ) that . EIF is a classical Kummer extension which is not a Kumrner extension with few roots of unity.
7. Give an example of a Cogalois extension which is not quasi-Kummer, and an example of a quasi-Kummer extension which is not Cogalois. 8. Give an example of a finite classical Kurnmer extension which is not Cogalois.
9. Give an example of a finite quasi-Kummer extension which is not a classical Kummer extension. (Hint: Use Theorem 7.4.3 and Exercise 3.)
R.EMARKS7.5.4. (1) For any real number field E, the extension ElQ is radical if and only if it is G-Cogalois for some group G. Indeed, any such . . , n f i ) / Q for some a l , . . . ,a, E extension has the form Q( rLfl,. which is Q" ( "*,. . . , ";/a,)-Cogalois. He already noticed in Exam. . . , " m ) / Q is Cogalois and ples 3.2.1 (1) that the extension Q( Cog(E/F) = Q" ( "fi,. . Q /). ,a . /,; "
"l&,
q,
(2) Any quadratic imaginary extension Q(&)/Q with d < 0 a squarefree integer is @ (&)-~ogalois, being a classical 2-Kurnmer extension.
(3) The determination of all imaginary algebraic number fields E which are G-Cogalois for some group G seems to be not so easy.
(4) The relations among the classes of extensions discussed up to this 0 point are given in Exercises 2 - 13.
10. Let n E N*. We say that a finite separable extension EIF is an n-G-Kneser extension if E/F is a G-radical extension with G / F * a finite group of exponent a divisor of n , such that p,(G) F * , and C4 E F whenever 1 E G. Prove that any finite separable n-G-Kneser extension is GKneser, but in general, a finite separable G-Kneser extension with G I F * a finite group of exponent a divisor of n is not necessarily n-G-Kneser.
+ c4

Give an example of an Abelian extension E/F with Gal(E/F) a bounded group of exponent n 2 2, such t h i ~ tgcd(n, e ( F ) ) # 1 and PTI(fl) F.
11. Show that the p2-pure extension considered in Exercise 3 is not p'- (Cp3)-Knew. 12. Show that any finite generalized n-Kummer extension E/F is an nG-Kneser exterision for some group G, but in general, not conversely. (Hint: Take F = Q(i) and E =
!z
~(z).)
Let F = Q(i) and E = F ( ~ ) Prove . that E/F is a generalized 8-Kummer extension, but not a classical Kummer extension. (Hint: Show that ,us(E) = { I , -1, i, -i} and $ E.) Show that for any odd prime p, the extension Q(Cp3 )/Q(Cp) is d pure, but it is not a generalized p"-Kummer extension. (Hint: Use Exercise 1, Chapter 3.) Give an example of a Cogalois Abelian extension which is not a classical Kummer extension. (Hint: Cse Exercise 3.)
13. Let n E W .Show that an n-G-Kneser extension is not necessarily n-pure, in particular, is not necessarily G-Cogalois. 14. Provide infinitely many examples of finite separable radical extensions E/F which are not G-Cogalois for any group G. (Hint: Use Corollary 7.4.5.) 15. (Tena [105]). Let n E W . Prove that there exist the following expressions by linear quadratic radicals:
Chapter 7
7.7 Hibliographical comments t o Chapter 7
189
for some ai E N*, Xi,pi E Q, i = 1,. . . , r , if and only if n E {1,2,3,4,6,8,12,24). (Hint: Use Corollary 7.4.5.) 16. (Barrera- Mora, Rzedowski-Calderdn, and Villa-Salvador [30]). Let EIF be a finite Galois Cogalois extension such that p ( E ) = p ( F ) . Show that EIF is a classical Kummer extension; in particular, E/F is an Abelian extension and Cog(E/F) E Gal(E/F). (Hint: Apply Proposition 7.2.4.) 17. Provide an example of a finite classical Kummer extension E/F which is Cogalois and p ( E ) # p ( F ) . 18. (Barrera-Mora, Rzedowski-Calderdn, and Villa-Salvador [SO]). Let F be an algebraic number field, and let E/F be a Galois Cogalois extension with r = Gal(E/F). Prove that there exists A a r such that A is Abelian and r / A 2 p ( E ) / p ( F ) . Furthermore, show that the extension E/Fix(A) is a classical Kummer extension. In particular, I? is a metabelian group. 19. (Barrera-Mora, Rzedowski-Calderdn, and Villa-Salvador [30]). Let E/F be an extension satisfying the assumptions of Exercise 18. Show that I ' is not necessarily the semidirect product of A and P(E)/P(F). 20. (Barrera-Mora, Rzedowski-Calderdn, and Villa-Salvador [30]). Let A be any finite Abelian group. Prove that for any given algebraic number field F there exists a finite extension EIF such that E/F is a Cogalois extension and Cog(E/F) E A. 21. (Barrera-Mora, Rzedowski-Calderdn, and Villa-Salvador [3O]). Let E/F be a Galois Cogalois extension with Galois group I?. Prove the following statements. , 6 F , then p I( I r l . (a) If p E P is such that C (b) If C4 $ F, then the 2-Sylow subgroup of r is isornorphic t o (Z2), for some r E W. 22. (Barrera- Mora, Rzedowski-Calderdn, and Villa-Salvador [SO]). Let F be any field, and let p E P be such that C, $ F . Prove that there is no Galois Cogalois extension E/ F with p 1 [ E : F 1. 23. (Barrera-Mora, Rzedowski-Calderdn, and Villa-Salvador [30]). Let r, n1, . . . ,n, E W* with n1 1 n2 1 . . . 1 n,, and let F be an algebraic
? , E F . Prove that there exist infinitely number field such that C, many Galois Cogalois extensions E/F such that
24. (Barrera-Mora, Rzedowski-Calderdn, and Villa-Salvador [30]). Let F be an algebraic number field, and let A be a finite Abelian group with A Z Z,,, and nl I n . (~ . . . I n,. Prove that the following two assertions are equivalent. (a) There exists a Galois Cogalois extension E/F such that
n : , ,
E
(b)
C,,,-,
F and F([,,?)/F is a pure extension.
25. Consider the following statements for an algebraic number field E . (a) E / Q is a Galois Cogalois extension. (b) E / Q is an Abelian Cogalois extension. (c) There exist r E N and square-free rational integers a,, . . . ,a, E 2 \ (0, -1, -3) such that a i / a j @ (1, -1, 3, -3) for all i # j in (1,. .. , r ) and E = Q(JG,... ,&). (b) true? (b) =+ (c). Is the implication (c) Prove that (a)
i
I
i i
Section 7.1. The concept of Kummer extension is a classical one in Field Theory. To distinguish such extensions from various generalizations, we refer to them as classical n-Kummer extensions (note that Artin [26], Karpilovsky [76], Lang [80], etc. refer t o them also as Kummer extensions of exponent n, and Bourbaki [40] as Kumnber extensions of exponent dividing n). All these sources provide a good account of the Kummer Theory. The idea t o place finite classical Kummer extensions in the framework of G-Cogalois extensions goes back t o Albu and Nicolae [19], and was also exploited by Albu [lo] and Albu and Tena [25] for infinite Kummer extensions (see Chapter 13). Section 7.2. Generalized Kummer extensions were introduced and investigated by Albu and Nicolae [19] for finite extensions, and by Albu and Tena [25] for infinite extensions.
190
Chapter 7
Section 7.3. The concept of Kummer extension with few roots of unity was introduced by Albu [3] for finite extensions, and by Albu and Tena [25] for infinite extensions. A systematic study of finite Kummer extensions with few roots of unity was undertaken by Albu [3]. Note that a series of results in Albu [3j can be deduced more directly, more simply, and more generally using the Kneser Criterion and the concept and properties of a G-Cogalois extension. Section 7.4. The notion of neat presentation was introduced and investigated by Greither and Harrison [63]. A neat presentation is a Galois extensions EIF of the following type: E = F ( x l , . . . ,x,), with Char(F) = 0 and xy,. . . ,xp E F for some n E N*, such that pP(R) C F for every p E 'P,,. Csing heavy cohomological machinery, which includes the LyndonHochschild spectral sequence, Greither and Harrison proved that any such extension is an extension with F* (xl, . . . ,x,)/F*-Cogalois correspondence, i.e., in our terminology, is an F*(xl , . . . ,x,)-Cogalois extension. Albu and Nicolae [19] introduced the more general concept of generalized neat presentation by dropping from Greither and Harrison's definition the condition " E I F is Galois" and by weakening the condition "Char(F) = 0" t o "gcd(e(F), n ) = I", and proved in a very simple manner (see also Theorem 7.4.2) that any such extension is still F*( x l , . . . ,x,)Cogalois. In this monograph, the generalized neat presentations were renamed quasi-Kummer extensions. Theorem 7.4.3 is due to Albu, Nicolae, and Tena [23]. A particular case of Corollary 7.4.4 for Galois Cogalois extensions was established by Barrera-Mora, R.r/,edowski-Calderbn,and Villa-Salvador [30] using the description of the Cogalois group of a Galois extension by means of crossed homomorphisms. The idea t o apply Theorem 7.4.3 t o prove Corollary 7.4.4 is due to Tena [105], to whom also belongs Corollary 7.4.5. Section 7.5. The fact that any finite Cogdois extension EIF is an extension with Cog(E/F)-Cogalois correspondence was first proved by Greither and Harrison [63j. The very short proof of this property provided in this section, b a e d on the n-Purity Criterion, is due to Albu and Nicolae 1191. Propositions 7.5.2 and 7.5.3, related to the dual of the Inverse Galois Theory Problem, are due to Barrera-Mora, Rxedowski-Ct-llderb~~, and VillaSalvador [30].
CHAPTER 8
G-COGALOIS EXTENSIONS AND PRIMITIVE ELEMENTS

By a well-known result, any finite separable extension E/F has a primitive element, i.e., an element u E E such that E = F ( u ) . However, in general, there is no practical procedure to find effectively such a u for a given extension E I F . The aim of this chapter is t o show that for a fairly large class of finite separable extensions, namely, the G-Cogalois extensions, it is possible t o provide a simple and easily manageable method of finding primitive elements. Section 8.1 contains the main result of this chapter (Theorem 8.1.2) which essentially establishes an easily verifiable necessary and sufficient condition for a finite sum of elements of G to be a primitive element of a G-Cogalois extension E / F . In particular, it follows that for any set of representatives S of the (finite) Kneser group G I F * of any G-Cogalois s is a primitive element of E I F . extension EIF, the sum CBES Applications of results of Sections 8.1 and 7.2 t o elementary Field Arithmetic are given in Section 8.2.
8.1. Primitive elements for G-Cogalois extensions Any G-Cogalois extension is by definition separable and finite, hence, by a classical result, it has in any case a primitive element. The main result of this section gives an easy way t o provide such an element: it can be obtained by summing up any elements X I , . . . ,x, of G such that G = F * ( x l , . . . ,x,) and 6 # 6 for every i # j in { I , . . . , n ) .
PROPOSITION 8.1.1. Let EIF be a G-Kneser extension, let n 2 2 be a natural number, and let X I , .. . ,x, E G be such that Zi# 6 for every i , j E { l , . . . , n), i # j . Then xl + . . . +x , , @ F .
192
Chapter 8
8.1 Primitive elements for G-Cogalois exter~sions
193
PROOF. Suppose that xl + . . . + x, = a E F. By Corollary 2.1.10, (21,. . . ,x,) is linearly independent over F , hence we cannot have a = 0. I t follows, again by Corollary 2.1.10, that there exists an i , 1 i n , such that Ei = 2 = 1, i.e., xi E F . Then X I + . . . + xi-1 + xi+l + . . . + X , E F , and, as above we deduce that xj E F* for some j E ( 1 , . . . ,n ) \ { i ) , i.e., 0 xi = 5j = 1, which is a contradiction.
h
< <
21,.
+ x,-1 E F ( H ) . By induction, we deduce that which shows that xl + . . ,%,-I E H , which finishes our induction. So, X I , . . ,x , E H , hence
G =Fa(xl, . . . ,x,) i.e., H = G . Then
< H < G,
0
THEOREM 8.1.2. Let E I F be a G-Cogalois extension, let n E W* , and let ( ~ ~ ) ~ < ibe < , a finite family of elements of G such that 6 # 6 for every i ,j E ( 1 , . . . , n ) , i # j. Then X I + . . . + x , is a primitive element of E I F if and only if G = F * ( x l ,... ,x,). PROOF. I f E = F ( x l +. . .+x,,), then E = F [ x ] where , x = X I + . . .+x,, hence any element of E , in particular any element g E G , has the form f ( x ) for some polynomial f E F [ X ] ,and consequently g = h ( x l , . . . ,x,,) for some polynomial h E F I X I , . . . ,X,]. Thus, g = hl + . . . + h ,,,, with hi E F * ( x l,... ,x,), and & # h; for every i , j E {1, . . . , m ) , i # j. I t follows that there exists i , 1 i rn, such that i j = h i , for otherwise, by Corollary 2.1.10, {g, h l , . . . ,h,) would be linearly independent over F , which is a contradiction. Hence g E F * ( x l ,. . . ,x,) for every g E G , i.e., G = F * ( x l , .. . ,x,).
and we are done.
COROLLARY 8.1.3. Let E I F be a G-Cogalois extension, let n E W , be a finite family of elements of G such that 6 # $ and let for every i # j in ( 1 , . . . , n ) . If x1 . . . + x, is a primitive element of E I F , then Xlxl + . . . + X,,x,, is also a primitive element of E I F for any family of elements of F*.
< <
PROOF. Set yi = Xixi for every i = 1,. . . ,n. Then F* ( X I , . . . ,x,) = # 6 for every i , j E { l , . . . , n ) , i # j . Now apply F * ( y l , . .. ,y,,) and 0 Theorem 8.1.2.
The next result shows that the condition "6 # 6 for every i { I , . . . ,n)" in the statement of Theorem 8.1.2 is not essential.
# j in
Row suppose that G = F* ( xl , . . . ,x,), where X I , . . . ,x , are as in the statement of the theorem, and consider the subextension F ( x l +. . .+x,)/F G and of E I F . By Theorem 4.3.2, there exists H such that F* H F ( x l + . . . + x,) = F ( H ) . W e are going t o prove by induction on n that
< <
I f n = 1, then F ( G ) = F ( x l ) = F ( H ) , and again by Theorem 4.3.2, we deduce that G = H , hence xl E H . Now assume that n 2 2 and the result holds for n - 1. W e are going t o prove that it also holds for n. Since X I + . . + x, E F ( H ) , we have
for some h l , . . . ,h , E H , with m 2 1 and hi # h j for every i ,j E (1,.. . , m ) , i # j. By Corollary 2.1.10, we deduce that { X I , .. . ,x,,) and { h l , .. . ,h,) are both linearly independent sets over F , and that {G,.. . , n { h l , .. . , h,,) # 0 . For simplicity, suppose that = h,. Then x , = Ah,, for some X E F * , and so, we have x , E H . Thus
h
COROLLARY 8.1.4. Let E I F be a G-Cogalois extension, let n E N * , and let be a finite family of elements of G . Consider a set of representatives { y l ,. . . ,y,) C { X I , .. . ,x,) of the subset { E , . . . , ) of G I F * . For each k , 1 6 k r , let Yk = { x i l E i = $ , 1 i n ) and zk = LEYk x . Then xl + . . . + x, is a primitive element of E I F if and onlyif G = F * ( { q l j J ~) ) , w h e r e J = { j I l , < j < r , z j # O ) . In particular, if { u l ,. . . ,u,) is any set of representatives of G I F * , then ul + . . . + u , is a primitive element of the extension E I F .
<
< <
PROOF. Clearly, CIqiGn xi = zk = X j E J zj and zj E G for every j E J , since zj = Xjyj for some X j E F*. Moreover, # for all 0 j # k in J . Now apply Theorern 8.1.2.
E E * . If F ( x l , . . . ,x n ) / F is an F* ( X I , .
. .
z)
LEMMA 8.1.5. Let E I F be an extension, let n E W ,and let X I , . . . ,x,, . . ,x,,)-Kneser extension, then the following assertions are equivalent. [ F ( x i )F : ]. ( 1 ) [ F ( x l,... , x , ) : F ] ( 2 ) The canonical morphism of F-algebras
=nr=l
191
Chapter 8
8.1 Prirriitive elements for G-Cogalois extensior~s
195
is a n isomorphism. ( 3 ) The family ( K ) , ... , of subgroups of F * ( x l , .. . , x n ) / F * is independent, i e . , if i l , . . . ,i,, E W, then
(z)
( 3 ) =+( 1 ) : The condition in ( 3 ) means precisely that ( . . . ,g) is the internal direct sum of its subgroups ( Z ).,. . , ( E l )(see 1.1.2), hence
c,
( Z , , xn)
h
(z) x ... x (c).

*
I
=
where dj = ord($) for every j , 1 j n. ( 4 ) If i l , . . . ,in E W, then xf' ...: x k E F* ==+ x: E F* for every k, 1 < j<n. ( 5 ) There exists a group isomorphism
< <
But, as observed in the implication ( 1 ) (5), d j = [ F ( x j ): F ] and [ F ( x 1 ,... ,x n ) : F ] = I F* ( X I , . . . ,x n ) / F * I. Consequently
[ F ( x l , . . , x n ) : F ] = 1 F * ( x l , .. . , x,)/F*
I ( Z , . . , x,) I
h
=I(%)x
. . . x ( ~ ) ~ = / ( ~ ) ~= ~d -l ... . . . d ~n ~ = ( n ~[ ) ~~ ( x I iF ) ]. i=l
(3) w ( 4 ) and ( 5 ) ===+ (6) are obvious, and (6) the last part of the proof of ( 1 ) (5).
=+
( 1 ) follows from
PROOF. ( 1 ) w ( 2 ) : The canonical morphism of finite dimensional

F-algebras
PROPOSITION 8.1.6. Let n E W * , and let X I , . . . ,x,, E R* be such that F ( x 1 , . . . ,x T 1 ) / Fis F* ( x l ,. . . ,~ ~ ~ ) - C o g a l o If zs.
[F(xl, -2,) : F
F ( x 1 ) @F . . . @F F(x,,) -+ F ( x 1 , . . . ,x,) is surjective, and it is injective if and only if [F(x~ QD ) F. . . @ F F(x,) : F ] = [ F ( x ~ . .,. ,x,,) : F ] ,
i.e., if and only if
]=
i=l
n
n
[ F ( x i ): F ] ,
then
F ( x l ,. . . ,x,,) = F ( x l + . . .
+ x,).
(1) (5): Since the extension F ( x l ,. . . ,x,,)/F is F* ( X I , . . . ,2,)Kneser, it follows that the extension F ( x j ) / F is K*( xj)-Kneser for every j, 1 j n by Corollary 2.1.10, hence
< <
PROOF. Without loss of generality we may assume that X I , . . . ,x , # F . Set E = F ( x l , . . . , x n ) , G = F* ( X I , . . . ,x,,), and di = ord(i$), i = 1, ..., n. By Theorem 8.1.2 we have only t o show that the cosets ( %) , . . . , (G) E R*/F* are mutually distinct. Assume that this is not the case, hence only r cosets among them are mutually distinct, with 1 r < n. Without loss of generality, we may suppose that these are G,. . . ,x,.
<
[ F ( x j ): F ] = ( F * ( x j ) / F * I = ord(6) = dj.
Csing again the fact tha.t F ( x l, . . . ,x,,)/F is F* ( X I , . . . ,x,)-Kneser, we have On the other hand, the group F* ( X I , ... ,x n ) / F * is isomorphic t o a quotient group of the direct product of groups (%) x . . . x Since both groups have the same order dl . . . . . d,, they are isomorphic. This imof subgroups of F * ( x l ,. . . ,x , ) / F * is plies that the family ( E ) .,. . , independent, and also proves the implication ( 5 ) (3).
The canonical group morphism

q : (Z) x ... x
(Z) -+ ( E...,q, ,
(cL).
(z)
-j,. , q ( ~ " , . . , -j, x , ) = ~ j .' . . . ex, is obviously surjective, hence the orders of these groups satisfy the inequality I(z,...,g)I I(%)I.....I(c)I=dl.....dr. But . . . , x,) = ( X I , . . . , x,) = G I F * , and so, the inequality above can be written as IG/F* 16 dl . . . . .d,.
(c,
<
198
Chapter 8
8 . 2 Applicatior~sto elerner~taryField Arithrr~etic(11)
199
w ) : =n:=, PROOF. The conditions pn(F (m,. . . , ;/a,)) s F and gcd ( n ,e ( F ) ) = 1 ensure by Theorem 7.2.3 that the extension F (m, . . . , * ) I F is F*( "fi, . . . , "*)-Cogalois. Now apply Proposition 8.1.6. 0
The example below shows that the condition in Corollary 8.2.6 on the degrees of the involved extensions is essential.
COROLLARY 8.2.6. Let n , r E W , let F be a field with gcd ( n ,e ( ~ ) =) 1, and let a l , . . . ,a, E F * . Suppose that p n ( F ( W , . . . , -1) E F and [ F ( m ,... , F] [ F ( % ) : F ] . Then F ( Z ,... , *) = F ( Z + . . . + &).
Then
i/ao E F
( w , . . . ,Cl/a,) if and only if there exist j , , . . . ,j, E
N and c E F* such that

PROOF. Set E = F @ , . . . , fi) arid G = F * ( m , .. . , AsBy Proposition 8.2.8, we deduce sume that i/ao E F ( m , . . . ,*). that = c. . .. . . ;/ii;'v
w).
m"
a
EXAMPLE 8.2.7. We have Q(45, Corollary 8.2.3, but
B, m)= Q ( J Z + fi+ m)by
for some j l , . . . ,j, E N and c E F * . Raising this equality t o the n-th . a:' . . . . @. power, we obtain a0 = cT1 Conversely, the equation in the statement of the corollary can be written as (%)'l= (c. z j l ...: @-$)rt If we denote
C = (c
m" . . . .
=
(C.
*h)-l
. a0 , then
./~
Kext, we present some applications to elementary Field Arithmetic of the property that generalized Kummer extensions are G-Cogalois (see Theorem 7.2.3 ( 1 ) ) .
By hypothesis, it follows that arid c -
C E F , hence
C)
fij'
. .. . -
PROPOSITION 8.2.8. Let r , no, n l , . . . ,n , E N* , and let ao, a l , . . . ,a, E F * . Let n = lcm ( n o ,n l , . . . ,n,), and suppose that gcd ( n ,e ( F ) ) = 1 and p n ( F ( " f i , . . . , "-)) C F, i.e., F ( " f i , . . . , ll*)/F is a generalized n- K u m m e r extension. if and only if there exist j l , . . . ,j, T h e n "@E F("&, . . . , "6) E N and c E F* such that PROOF. Set E = F ( " f i , . . . , "fi) and G = F*("*,. . . , ";/a,). By Theorem 7.2.3, E I F is a G-Cogalois extension. Now, observe that , , ( E ) E pT,(E)5 F . So, we can apply Lemma 6.1.6 since no I n , we have p to deduce that E F ( " f i , . . . , ";/a,) if and only if "@5j E G, as desired. 0
C E F.
This shows that
E F ( m , . . . , %).
COROI,I~ARY 8.2.10. Let r , no, nl ,. . . n , E N* , and let ao, a1 ,. . . a, E "$Gj can be written as a finite s u m of monomials of form QL c . "fi ' . . . . " f i j r 7 with j l , . . . ,j, E N and c E p, if and only if is itself such a monomial.
"w
",""
PROOF. The condition about the possibility to express "fi as a sum of monomials of the indicated form means precisely that "fi E Q( "fi, .. . , Since pm (Q( . . . , "6)) { -1, I } for every m E W , we can apply Proposition 8.2.8 to obtain the desired result.
"s).
"m,
Tm
We will now examine the elementary arithmetic of finite G-radical extensions E I F with e x p ( G / F * ) a prirne number p; in other words, of finite pbounded radical extensions.
COROI,I,ARY 8.2.9. Let r , n E W , let ao,al,. . . ,a, E F * , and suppose that gcd ( n ,e ( F ) ) = I , and P ~ ~ ( F ( ; ~ / . O , . . . ,*)) {-1,l) or pTL(R) C_ F, i n other words, the extension F ( m , , . . . ,w ) / F is either a n n - K u m m e r extension with few roots of unity or a classical n -K u m m e r extension.
LEMMA 8.2.11. Let F be a n arbitrary field, let p > O be a prime n u m ber, other than the characteristic of F , let r E W ,let a l , . . . ,a, E F * , and let ,. . . ,fiE R denote any fixed p-th roots. Assume that
200
Chapter 8
8.2 Applications t o elernerltary Field Arithmetic (11)
20 1
Then, we have either C, E F or p P ( F ( m ,. . . ,fi)) = {I), in other words, F , . . . , G )/F is either a classical p-Kummer extension or a p-Kummer extension with few roots of unity.
(m
PROOF. Since pp ( F ) = p,(R)n F* is a subgroup of the group pP(R) of order p, we have either p, ( F ) = pp ( a ) or pp ( F ) = {I). If pp( F ) = pp (R) then clearly Cp E F . Now assume that p,(F) = (1). We are going t o prove by induction on r that for any field K with e ( K ) # p and p p ( K ) = {I), any {bl,. . . , b,) R* such that b r E K for all i, 1 i r , and
\,
< <
(3) By Exercise 3, a result similar t o that of Lemma 8.2.11 is valid without the condition "[ F , . . . ,&) : F ] = pr " , but only for spe. . ,- E 0. This result involves the condition cific p-th roots fi,. Cl (n; a l , . . . ,a,), which extends the condition Cl (n; a ) defined in Section 1.3. Let n , r E W ,and let a ] , . . . ,a, E F * . We say that the field F satisfies the condition Cl (n; a l , . . . ,a,) if for every i , 1 i r, the binomial ,that pn ( F ,. . . ,fi)) c X"-ai possesses a root in 0 , say 6 such {-1,l). If gcd(n, e ( F ) ) = 1, then F . . . ,* ) I F is a finite nKummer extension with few roots of unity. Some properties of fields satisa1 ,. . . ,a,) are stated in Exercises 5 - 8. 0 fying condition Cl (n:
(m
< <
(m,
(m
PROPOSITION 8.2.13. Let F be an arbitrary field, let p > 0 be a prime number, other than the characteristic of F , let r E N* , let a1 , . . . ,a, E F * , and let . . . ,& E R denote any fixed p-th roots. If
For r = 1 observe that we cannot have bl E K since [ K ( b l ) : K ] = p 2 2. Hence bl $ ! K . Assume that pP(K(bl))= pp(R). Then Cp E K(bl), i.e., K ( & ) K(bl), and consequently,
m,
then the following assertions hold.

But [K(Cp) : K j p(p) = p-1 and [ K ( b l ) : K ] = p, hence necessarily we must have [ K (Cp) : K ] = 1, i.e., Cp E K . Then, Cp E pp(R)nK = p p ( K ) = {I), which is a contradiction. Thus pP(K(bl)) is a proper subgroup of the group pP(R) of prime order p. This implies that /l,(K(bl)) = {I), as desired. Now assume that the assertion is true for a given r E N*, and prove it for r + 1. So, let {bl,. . . ,b,, b,+]) C R* with by E K for every i , 1 i 6 r 1, and [ K ( b l , . . . ,b,, b,+l) : K ] = P,+'. We have t o show that pp(K(b17 - - . br, br+l)) = {I). Set E = K ( b l , . . . ,b,). The equality [ K(b1,. . . ,b,, b,+~): K ] = pr+' clearly implies that [E(b,+1) : E ] = p and [ E : K ] = p'. By the step r of the induction, we have p,(E) = (1). Now, by the step r = 1 of the induction applied to the field E and-the element b,+l we deduce that C1 ) (1). /lp(E(br+l)) = (1) i.e.7 /hp(K(bl . ,br b r + ~ ) =
-
<
<
(1) The extension F ( is F * ( $ K , . . . , f i ) , . . ,F Cogalois. (2) The Kneser group F* ( ,. . . , m ) / F * of the extension E/F is isomorphic to the direct product Z i of r copies of the group Z,. (3) 1 . . , &) I = /(a; ,. . . ,a^,)I = p P , where ii denotes for any a E F* its coset in the group F*/F*P. (4) If il , . . . ,i n E N and af' . . . . . a? E F*P, then p 1 i l , . . . ,p 1 i,. =F(fi f i ) . (5) F ( m ,
m
h
(s,.
...,rn)
-$ a
PROOF. (1): By Lemma 8.2.11, the extension F ,i f ( .. ., f i ) / F is either a classical p-Kummer extension or a pKurnmer extension with few roots of unity, hence a generalized pKummer extension, and so, by Theorem 7.2.3 ( I ) , it is F* . . . , -)-Cogdois.
(m,
R.FXARKS8.2.12. (1) Lernma 8.2.11 trivially holds for p = e ( F ) since /do) ={W (2) Exercise 2 shows that condition " [ F ( m , . . . ,&) Lemma 8.2.11 is essential.
:
(2) By ( I ) , E/F is in particular an F*( fi, . . . , & )-Kneser extension. Now apply Lernrna 8.1.5.
(3) follows from Theorem 7.2.3 (3), and (4) follows from Lemma 8.1.5.
(5) follows from (1) and Proposition 8.1.6.
F ] = p' " in
202
Chapter 8
8.2 Applicatior~sto elerr~er~tary Field Arithrr~etic(11)
203
REMARK 8.2.14. If the condition " [ F ( m , . . . ,-) : F ] = p"' in Proposition 8.2.13 is not fulfilled, then the result may fail. Indeed, the extension Q(Y3, G fi)/Q has degree 6, but the subgroup (2, 2 ) of Q /p3 has order 3. 0
n n
Since u is a root of the polynomial XP -a, it follows that f { XP - a, hence f is a product in n[X] of m binomials of form X - $u, hence
for some r E N. Then
COROLLARY 8.2.15. Let p be a prime other than the characteristic of a field F . Let E = F ( u ) with u E fl any root of an irreducible polynomial XP - a E F[X]. Then an element v E E satisfies vp E F if and only if v = bun for some b E F and n E N. PROOF. If v = bun then vp = bPan E F . Conversely, assume that vP = c E F . Clearly, we may assume that v # 0. Since [ F ( u ) : F] = p, EIF is an F*(u)-Cogalois extension by Proposition 8.2.13 ( I ) , and pP(E) C F by Lemma 8.2.11. Hence, by Lemma 6.1.6, we deduce that v E F*(u), i.e., v has the desired form. 0 COROLLARY 8.2.16. Let p be a prime other than the characteristic of a field F. Let u, v E R be any roots of the irreducible polynomials XP - a and XP - b E F [ X ] , respectively. Then [ F ( u , v) : F] = p' unless b = cpan for some c E F and n E N . PROOF. If b 6 F(u,)p,then the polynomial XP - b remains irreducible over F ( u ) by Lemma 1.3.2, hence [ F ( u , v ) : F] = p'. If b E F(u)P, then b = WP for some w E F ( u ) , so w = curl for some c E F* and n E N by Corollary 8.2.15. Raising this equation t o the p t h power, we obtain b = cpan. COROLLARY 8.2.17. Let p be a prime other than the characteristic of a field F . Let u, v E R be any roots of the irreducible polynomials XP - a and XP - b E F[X]. I f [ F ( u , v) : F ] = p2 then F ( u , v ) = F ( u v).
Thus uC,Ps = at(&bo)' E F, and so, a = UP = (uCpPs)P E FP, as desired.
PROPOSITION 8.2.19. Let F be an arbitrary field, let p > 0 be a prime number, other than the characteristic of F , let r E N,r 3 2, let a l , . . . , a , E F * , and let m , . . . ,fiE fl denote any fixed p-th roots. Further, let denote E = F ( m , . . . ,ga,_l).Then either [E(@) : E ] = p, or
for some j l , . . . ,j, E N and c E F*.
PROOF. If a, @ E" then the polynomial XP - a, is irreducible in E [ X ] by Lemma. 1.3.2, so
Now, assume that a, E EP. Then a, = @ f o r some u E E = F ( m , . . . ,q a x ) . If we set K = F (Cp), then UP E K and the ex. . . ,va,_l)/K is a classical pKurnmer extension, so it is tension K (=, ,.) .l._,,a v . By G-Cogalois by Theorem 7.1.6 ( I ) , where G = K* (fi Lemma 6.1.6, it follows that u E G, i.e., for some v E K* and j l , . . . ,j,-l E N. Raising the last equation t o the p t h power, we obtain
PROOF. Apply Proposition 8.2.13 (5).
0
-.i1 . i r 1 E Kp n F. But [ K : F] = . . . . a,-, This implies that vp = a, . a, [F(CP) : F] 6 p - 1, hence UP E F P by Lemma 8.2.18, i.e., vp = cp for some c E F*.Thus a, = CP . a:' . . . . . r-1 7
LEMMA 8.2.18. Let F be any field, let p > 0 be any prime number, and let K / F be a finite extension such that [ K : F ] < p. Then KP n F = FP.
PROOF. \Ye will proceed as in the proof of Lemma 1.3.2. The inclusion K P n F is obvious. Kow, let a E KP n F , i.e., a = UP E F for some u E K . Let f = X"' + brrb-lXrrL-l+ . . . + blX + bo be the minimal polynomial of u over F . Since F ( u ) E K and [ K : F ] < p, it follows that 1 m < p, hence gcd(m, p) = 1. Then 1 = ms + pt for some s, t E Z .
F P
and we are done.
<
Chapter 8
8.4 Hibliographical cornmer~tst o Chapter 8
205

1. Does the result in Theorem 8.1.2 hold if one weakens the hypothesis "E/F is a G-Cogalois extension" to " E/F is a G-Kneser extension" ? 2. Show that if condition "[ F ( m ,. . . ,m ) : F ] = p"' in Lemma 8.2.11 is not satisfied, then, in general, the result does not hold for any choices of p t h roots . . , f i .(Hint: Take p = 3, r = 2, F = Q, a1 = a;! = 2, and consider the 3-th roots of unity fi, fiC3.) 3. Let F be an arbitrary field and let p > O be any prime number. Prove that for any r E W and any a l , . . . , a , E F*, either there exist specific p t h roots . . . ,& E R such that . . . ,m ) ) = {I), (i.e., the field F satisfies condition pp(F C(p;ao,... ,a,)), or CP E F .
9. (Zhou [113]). Let r, nl ,. . . ,n, E W , let pl , . . . ,p, E P be distinct primes, and let xk E @ be any root of the polynomial XTLk - ppk, k = 1 , . . . ,r. Prove the following statements. (a) [Q(xl ,. . . ,x,) : Q ] = nl . . . . .n,. (b) Q(x1,. . . 2,) = Q(x1 + . . . + x,).
10. (Zhou [113]). Let r, m, nl ,. . . ,n, E W , and let pl , . . . ,p, E P be distinct primes such that pk I m, m = p p m k , gcd(mk, p k ) = 1, and gcd(nk,p; - p2-1) for every k = 1 , . . . r. Let xk E C be any root of the polynomial Xnk - ppk, k = 1,. . . ,r. Prove the following statements. (a) [Q(Crn)(x~,... ,x,) :Q(Crn)] = n 1 . . - . . n , . (b) Q(CrrJ(x1,. . . ,x,) = Q(Crn)bl + .. + 4 .
e,.
(m,
m,
4. Let F be any field, let p > O be any prime number, let a E F*,and suppose that Cp E F and @ F. Show that [ F ( ) : F ] = p. 5. Let F be an arbitrary subfield of R. Show that F satisfies the condition C1(n; a , , . . . ,a,), where r E W* is arbitrary, and, either n E N* is an arbitrary odd number and a ] , . . . ,a, E F* are arbitrary, or n E W is an arbitrary even number and a l l . . . , a, E FS are arbitrary.
6. Show that a field F satisfying the conditions C1(n; a l ) and Cl (n; a;!) may not satisfy the condition Cl (n; a ] , az).
11. (Vdez [108]). Let F be an arbitrary field, let n E N* be such that gcd(n, e ( F ) ) = 1, and let XT' - a , XrL- b be irreducible polynomials in F [ X ] with roots u, v E R, respectively, where R is an algebraically closed overfield of F. If the extension F ( u ) / F has the CSP, then prove that the following assertions are equivalent. (1) The fields F ( u ) and F ( v ) are F-isomorphic. (2) There exists c E F and j E N with gcd(j, n ) = 1 and a = bJcn. (Hint: Use Lemma 6.1.6 and Exercise 17, Chapter 4.)
12. Let F be an arbitrary field, and let n E N* be such that CT, E F and gcd(n,e(F)) = 1. Let X" - a, XT" b be irreducible polynomials in F [ X ] with roots u, v E R, respectively, where R is an algebraically closed overfield of F . Prove that the following statements are equivalent. (1) F ( u ) = F(v). (2) There exists c E F and j E N with gcd(j,n) = 1 and a = PC''. (Hint: Use Exercise 11 and Theorem 7.1.6 (I).)
7. Let n E W be an odd number, let r E N*, a l , . . . ,a, E F*,and suppose that F satisfies the condition Cl (n; a ] , . . . ,a,). Prove that if EIF is any Abelian extension with Cn E L, then
: F] = [ L ( m , . . . ,-) [ F ( m , .. . ,=)
:L],
where 6denotes for every i = 1 , . . . ,r a specified root in R of . . ,s ) ) {-1,l). the polynomial XrL-ai such that p r L ( F
(m,.
8. Let n E W be an odd number, let r E W , a l , . . . ,a, E F*,and suppose that F satisfies the condition Cl (n; a l , . . . ,a,). Prove that , . . . ,& ), then if E/ F is an Abelian extension with E E F E = F . As above, 6denotes for every i = 1 , . . . ,r a specified root in R of the polynomial X n - ai with p,,(F (*, . . . , ;/a,)) {-Ll}.
(m

Section 8.1. The results of this section are due to Albu and Nicolae [20]. Section 8.2. The results of this section are due t o Albu [3], [7], [12], [13], Albu and Nicolae [20], Albu, Kicolae, and Tena [23]. Some of the
200
Chapter X
original results of Albu i3] and Albu, Kicolae, and Term [23] presented in this section were either incorrect or had incomplete proofs and we have made the necessary changes. To the best of our knowledge, Proposition 8.2.1, Corollary 8.2.2, Corollary 8.2.3, as well as well as a si~nilarresult for finite classical Kummer extensions, were not known before 1995, when the paper of Albu and Kicolae [20] appeared. Corollary 8.2.5 is folklore, but without using Proposition 8.2.1, its proof is not a t all a brief as ours. Corollaries 8.2.15, 8.2.16, 8.2.17 appear in Kaplansky [74], but were originally proved in a different manner. Proposition 8.2.19 is essentially due to Baker and Stark [28], who proved it only for algebraic number fields. Our proof, b a e d on Lemma 8.2.18 and taken from -4lbu [7j, is valid for more than just algebraic number fields.
CHAPTER 9
APPLICATIONS TO ALGEBRAIC NUMBER FIELDS

Cogalois Theory has nice applications t o algebraic number fields. This chapter presents some of them. After reviewing, in Section 9.1, the standard notation, notions, and facts on algebraic number fields which will be used in the sequel, we show, in Section 9.2, that some classical results due t o Hasse, Besicovitch, Mordell, Siege1 are immediate consequences of basic facts in Cogalois Theory. In Section 9.3 we present a surprising application of the Kneser Criterion in proving a very classical result in Algebraic Number Theory claimed by Hecke (but not proved) in his book [71] and related t o the so-called Hecke systems of ideal numbers.
9.1. Number theoretic preliminaries

Below, we present some of the concepts and results from Algebraic Number Theory which will be used in the next sections of this chapter. All rings considered in this section are supposed t o be commutative with nonzero identity element. Throughout this section R will always denote such a ring.
9.1.1. INTEGRAT, EXTENSION

I
Integral element. Let R be any ring, let S be an overring of R (that is, a ring containing R as a subring), and let u E S. Then u is said t o be an integral element over R if there exist n E W* and elements ao,. . . ,a,,-, in R such that u satisfies the equation
urL a,-lun-'
+ . . . + a0 = 0,
208
Chapter 9
9.1 Number theoretic preliminaries
209
in other words, if u is a root of the monic polynomial If E/F is a field extension, then u E E is integral over F if and only if u is algebraic over E . Ring extension. If S is an overring of a ring R , then the pair (R, S ) is called a ring extension, and in this case we shall write S I R . If S I R is a ring extension, then the underlying additive group (S, +) of the ring S may be viewed as a nodule RS via the scalar multiplication R x S -+ S, (a, x) I+ax, where a x is the product of the elements a E R and x E S under the given multiplication on S. In fact, this structure of R-module on S together with the given ring structure on S endow S in a natural way with a structure of an R-algebra. A ring extension S / R is said to be finite if RS is a finitely generated module. Integral extension. We say that a ring extension S I R is an integral extension if every element of S is integral over R. A field extension is integral if and only if it is algebraic. As is known, any finite field extension EIF is necessarily an algebraic extension. The corresponding result for ring extensions is the following: any finite ring extension is an integral extension. Integral closure. If S I R is a ring extension, then the set Rk of all elements of S which are integral over R is called the integral closure of R in S . The integral closure Rk of R in S is a subring of S that includes R. The ring R is said t o be integrally closed in S if Rk = R. The integral closure Rk of R in S is integrally closed in S , and Rk/R is an integral extension. A domain is a ring without zero divisors, that is, a ring having the following property: whenever a , b E R and ab = 0, it follows that a = O or b = 0 . An integrally closed domain is a domain which is integrally closed in its field of quotients. Algebraic integer. An algebraic number field, or simply, a number field is any subfield K (which automatically contains Q as a subfield) of the field C of complex numbers, such that the extension K I Q is finite. The positive integer [ K : Q ] is called the degree of the number field K . Let K be an algebraic number field. The integral closure Zk of Z , which is a subring of K , is called the ring of algebraic integers of K , and
is denoted by DK. Observe that Dq = Z , and this is the reason why the elements of Z are called rational integers. An algebraic number (resp. algebraic integer) is any z E (C which is algebraic (resp. integral) over Q (resp. over Z). Quadratic integer. A number field K is said to be quadratic if [ K : Q] = 2. A subfield K of C is a quadratic number field if and only if there exists a square-free integer d E Z\ {I}, which is uniquely determined, such that K = Q(&), that is, K = { a b& 1 a , b E Q}. A number s E Z is said t o be square-free if p" I( s for every p E P. For any square-free integer d E Z\ {I},the ring DQ(&) of integers of the quadratic field Q(&) has the following form:
Definition and examples. Let R be a domain with field of quotients K. A fractional ideal of R is any R-submodule I of R K having the property that there exists an a E R , a # O such that a1 2 R , where a I = { a x I x E I}. If I is a fractional ideal of R, then I 5 R if and only if I is an ideal of the ring R , and we will then say that I is an integral ideal of R. Any cyclic submodule R x of R K , with x E K , is a fractional ideal of R , and we call it a principal fractional ideal of R. More generally, any finitely generated R-submodule of R K is a fractional ideal of R. Operations on fractional ideals. If I and J are fractional ideals of R , then I + J, I n J , and I . J are also fractional ideals of R , where
is the product of I and J. A fractional ideal I of R is said to be invertible if there exists a fractional ideal J of R such that I ,J = R. The set FR of all nonzero fractional ideals of a domain R is a commutative monoid with respect t o the multiplication of fractional ideals, having R as identity element. We will see in the next subsection that the monoid ( f i ,) is a group precisely when R is a Dedekind domain.
9. 1 Number theoretic prelimir~aries
Examples.
such that PID. A principal ideal domain, abbreviated PID, is a dornain is a principal ideal, i.e., I = na for some a E R. every ideal I of
Prime ideal. Let R be a ring. -4n ideal P of R is said t o be prime if P # R and for any a , b E Il such that ab E P it follows that a E P or b E P. The zero ideal O of R is a prime ideal if arid only if R is a dornain. Prime element. Let I2 be a ring. An element p E R is said t o be prime if p # U ( n ) and for any a , b E R such that p 1 ab it follows t h a t p ( a or p 1 b, or ecluiva.lently, if the principal ideal n p generated by p is a prime ideal. UFD. -4 domain R is said t o be a unique factorization domain, abbreviated U F D , if every element a E R \ ( C ( I 2 )CJ (0)) factors into a product of prime elements. The factorixatioli is necessarily unique up t o unit multiples and the order of the fxt,ors. .41ly PID is a CFD. Maximal ideal. .4n ideal Ad of a ring R is said t o be nmxinml if A1 # R and whenewr I is an ideal of R such that A l C I. then I = Ad or I = R. Noetherian ring. A ring R is Noetherian if every ideal of I? is finitely generated, or equivaiently, if every ascending chain of ideals
( 1 ) -4ny field is a Dedekind domain. (2) Any PID is a Dedekind domain. In particular, the ring Z of rational integers, and the ring F [ X ] of polynomials in one indeterminate over any field F are Dedekind domains. (3) A Dedekind domain R is a L F D if and only if R is a PID. (4) Let R be a Dedekind domain, let K be field of quotients of R , and let L be a finite separable extension of K. Then, the integral closure R), of R in L is a Dedekind domain. In particular, for any algebraic number field I<, the ring D = Zk of algebraic integers of K is a Declekirid domain. Ideal class group. Let R be a Dedekind domain with field of quotients K . The set pH = { R x I x E K* ) of all nonzero principal fractional ideals of R is a subgroup of the group FR of a11 nonzero fractional ideals of R, so it makes sense t o consider the quotient group
I , g I,
of
I3
5 ...
E
which is called the ideal class group of R. The order hR E N* U {oo) of t,he group C f H is called t h e class number of R. The Dedekind domain R is a PID if and only if hH = 1.
I? is eventually constant, i.e., there exists an no I,,+ for all n 3 no.
N* such t h a t I,, =
Throughout this subsection K will denote an algebraic number field and 0 K the ring of integers of K.
Characterization. The following statements are equivalent for a domain n. (a) R is a Koetherian integrally closed domain and every nonzero prime ideal of I? is maximal. ( b ) Every nonzero fractional ideal of R is invertible, in other words, the monoid ( F H.) , of nonzero fractional ideals of R is a group. (c) Every integral ideal I of I2, I # A?, is representable as a product of finitely many prime ideals of I?. -4 domain I? satisfying one of the equivalent con~lit~ions above is called a Dedekind domain.
Group of units. The group of units U ( D K ) is called, by abuse of terminology, the group of units of K , and is denoted by UK. Root of unity. Since K C @, for any n E N* we c i ~choose l as a primitive n-th root of unity over K the complex number
T h e group p ( K ) of all roots of unity in K is a finite cyclic group.
Decomposition in prime ideals. Since DK is a Dedekind domain, any ideal I of D K , 0 # I # DI(, is representable as a finite product of prime ideals of Dl(. Moreover, this representation is unique up t o the order of prime ideals in the representation.
21 2
Chapter 9
9.2 Some classical results via Cogalois 'Theory
I
I
Ideal class group. The group FDK (resp. PD,) of all nonzero fractional ideals (resp. principal fractional ideals) of D K is called, by abuse of terminology, the group of nonzero fractional ideals (resp. the group of nonzero pr2ncipal fractional ideals) of K , and is denoted by .FK (resp. P K ) . The ideal class group CIDK := FK/PKis called, also by abuse of terminology, the ideal class group of K , and is denoted by CIK. The elements of CIK are called ideal classes of K . A famous theorem due t o Dirichlet states that elK is a finite group. The order of CtK is denoted by h K ,or simply by h when there is no danger of confusion, and is called, by abuse of terminology, the class number of K . Extension of ideals. Let L / K be a field extension of algebraic number fields, and let I E FI,. Then D L / D K is a ring extension, and Dl, is the integral closure of DK in L. For any I E FKset
PROOF.Clearly, K ( x l , . . . ,x,)/K is a classical n-Kummer extension, so it is a G-Cogalois extension by Theorem 7.1.6 ( I ) , where G = K ( X I , . . . ,x, ). In particular, K ( x l , . . . ,x,)/K is a G-Kneser extension, hence Now apply Lemma 8.1.5 to deduce that
Then IDr, E .TI, and
~~
9.2. Some classical results via Cogalois Theory
The aim of this section is t o present some classical results due t o Hasse, Besicovitch, Mordell, and Siege1 as easy consequences of basic facts from Cogalois Theory. Throughout this section K denotes a fixed algebraic number field, and for any a E K * and n E N*, @ designates a root (which in general is not specified) in C of the polynomial Xn - a E K [ X ] . However, if K is a subfield of the field R of real numbers and a > 0, then @ will always mean the unique positive root in R of the polynomial XTL - a.
The fact that K ( x l , . . . , x,)/K is a G-Kneser extension can be shown more directly as follows, using only the Kneser Criterion. Let p E P. If p 1 n , then we have CP E K , since [ , E K by hypothesis. Now assume that p n. Then up bn = 1 for some a , b E Z . Observe that G7' E K * , so, if b E G for an odd prime p, then ( : E K . It follows that - [up+bn - ([: ) % K , as desired. PP Similarly, 1 C4 E G implies that (1 <4)n E K . Since (1 + [4)4 = -4 E K , and we are assuming that 4 1 n, we deduce as in the proof of the Kneser Criterion (Theorem 2.2.1) that the element 1 C4 E G / K * has order 1 or 2, which implies that E K . By the Kneser Criterion, it follows that K ( x l , . . . ,x,)/K is a G-Kneser extension. 0
THEOREM 9.2.2 (SIEGEL, 1972). Let K be an algebraic number field, let r E N*, let n l , . . . , n , E N*, and let X I , .. . ,x, E C be such that xLk E K for all k, 1 6 k r . Assume that either c,,, E K for all k, 1 k 6 r , or K C R and xk E R$ for all k, 1 k 6 r. Then
<
<
<
THEOREM 9.2.1 (HASSE,1930). Let K be an algebraic number field containing C,, for. some n E W , let r E W , and let X I , .. . ,x, E C* be : E K for all k, 1 k 6 r . Assume that the following condition such that x is satisfied:
<
PROOF. Set G = K ( x 1 , . . . ,x, ) and L = K ( x l , . . . ,x,). Observe that L / K is a separable G-radical extension, and so, the desired equality means precisely that L / K is a G-Kneser extension. If K 5 R and x k E R$ for all k, 1 6 k r , then L 5 R, hence L / K is clearly a pure extension (see also Examples 3.1.4 (I)), and thus, it is G-Kneser by Lemma 3.1.6.
<
(t)
7711,.
. . , m,
N and XI'"
. . . . - x,
Trl
xTk E K , k = 1 , . . . ,r.
Now, assume that C,, E K for all k, 1 k r , and let n = l c m ( n ~.,. . ,n,). Then n = nkbk for some bk E Z, 1 6 k r. Clearly,
< <
<
211
Chapter 9
9.2 Sorrie classical results via Cogalois Theory
215
gcd(bl,. . . ,b,) = 1, hence 1 = bkck for some c l , . . . ,c, E Z. It follows that -(~2)" . .. . . E K, L = b l c l f ...+b,.c,. is a primitive nk-th root of unity for all k = since C,, E K and 1 , . . . ,r . By the proof of Theorem 9.2.1, L / K is a G-Kneser extension.
cTL
(~2)'~
Since the decomposition of a into a product of positive prime integers is unique up to the order of the factors, we deduce that r = s and there exists a permutation a E 6, such that pj = q,(j) and kj = ni,(j) for all j = 1 , . . . ,r , hence n 1 kj for all j = 1 , . . . ,r .
(3) ==+ (2) and (2) ===+ (1) are obvious.
[?
THEOREM 9.2.3 (MORDELI,, 1953). Let K be an algebraic number field, let r E N*, let n l , . . . , n , E N*, and let X I , .. . , x, E C be such that x i k E K for all k, 1 k r , and
COROLLARY 9.2.5. Let T E N,r 2 2, and let a l , . . . ,a,, n l , . . . , n , E N* be such that for every k, 1 6 k 6 r , and every Zk, 1 lk < izk one has
<
<
( " f i ) ' k
$ !
(tt)
m l , . . . ,m, E
N and x r ' . . . . x r r K
===+ n k 1 m k , k = 1 , . . . ,r.
(*I
Then
N. Assume that the following condition is satisfied. a , , . . . ,ak are relatively prime in pairs.
[Q(''&
Assume that either E K for all k, 1 for all k, 1 6 k r. Then
<
c,,
< k 6 r , or K E IW and xk E IW;

...:
n,.
, . . . ,";/a,) : Q ] = n l ...: n,.
[ K ( x l,..., 2,): K ] = n l
PROOF. We are going t o show that condition (*) implies condition ( t t ) in Theorem 9.2.3. Let m l , . . . ,m, E N be such that If n = lcm(nl , . . . ,n,), then n = niki for all i = 1 , . . . , r , hence the relation above can be written a s Since ai 3 2 for all i = 1,. . . ,r , we can factor each ai into a product of prime numbers. Let a l = q:' . . . . - qi3, with s , j l , . . . ,js E W and ql , . . . ,q, different positive prime integers. According to condition (*), any of the prime numbers ql, . . . ,q, does riot occur in the factorimtions of a2, . . . ,a, into products of prime numbers, so, by Lemma 9.2.4, we deduce that n I k l m lji, hence nl I m l ji for all i = 1,. . . , s. Consequently, ayl = q y . .~ . . - q?ljs is an n l - t h power of some natural number, i.e.,
PROOF. The extension K ( x l , . . . ,x,)/K is K ( X I ,. . . ,x, ) -Kneser by Theorem 9.2.2. To conclude, observe that condition ( i t ) is the same as condition ( t ) in Theorem 9.2.1. Now, apply Lemma 8.1.5 t o obtain the 0 desired equality. LEMMA 9.2.4. Let r, n, k l , . . . ,k, E N*, and let p l , . . . ,p, be diflerent positive prime integers. Then, the following statements are equivalent.
k (2) yp:l . . . . pTr. E N. (3) n ( kj for all j = 1 , . . . , r .
PROOF. For simplicity, denote a = pfl . . . - p f r and x = @. (1) =j (2): Assume that x E Q. Observe that x is a root of the moriic polynomial XrL- a E Z[XI, i.e., x is an integral element over Z. I t follows that x E Z since Z is integrally closed. But x > 0, hence necessarily x E N.
(nw)7'L! E N. But, by Lemma 9.2.4, the condition ( ' L s ) l l $ N for every

Another consequence of Theorem 9.2.3 is the following classical result.
11, 1 6 I I < n l means exactly that ord( "fi) = nl in the group R* /Q . Thus, we must have n l I ml. In a similar way, one shows that ni ( nzi for all i = 2,. . . , r , and we are done. D
(3): Assume that x E N. Since x 2 2 and Z is a UFD, we (2) can factor x as x = qi' . . . . . q i/ , where s , i l , . . . , is E N* arid ql , . . . ,q, are different positive prime integers. This implies that
THEOREM 9.2.6 (BESICOVITCH, 1940). Let r E N', let p l , . . . ,p, be different positive prime integers, let bl, . . . ,b, E N' be not divisible by any of these primes, and let a , = b l p l , . . . ,a, = b,p,. Then, for any nl , . . . , n, E N* one has
[ Q ( " G , .. . , '';/a,): Q ] = n , . . . . n,.
,
216
Chapter 9
9.2 Some classical results via Cogalois Theory
21 7
PROOF. We will show that condition ( t t ) in Theorem 9.2.3 is satisfied. Let m l , . . . ,m, E N be such that y & rial . m.,E Q.
0 . .
(4) Necessary and sufficient conditions for an arbitrary finite radical extension F ( x l , . . . , x , ) / F to have degree nl . . . .an,, where X I , .. . ,x, E R* are such that xYi E F* for every i , 1 i r, are provided in Exercise 8. 0
< <
:
If n = lcm(nl,. . . ,n,), then n = niki for all i = 1 , . . . , r , hence the rehtion above can be written as By hypotheses, pl , . . . , p r are different positive prime integers and bl , . . . , b, are not divisible by any of these primes. By Lemma 9.2.4, we deduce that i for all i = 1 , . . . ,r , as desired. n I kimi, hence ni / m 0 COROLLARY 9.2.7. Let r E W , let p l , . . . ,p, be different positive prime integers, and let n l , . . . ,n, E N* be arbitrary. Then
E X A M P L9.2.9. E We are going t o calculate the degree [Q(y2+
m) Q]
and t o exhibit a vector space basis of the extension Q ( Y E To do this, we first apply Corollary 8.2.3 t o deduce that ~ ( 4+3 Clearly hence
+m ) / Q .
m)= ( ? ( Y E , m).
PROOF. Apply Theorem 9.2.6 for bl = . . . = b, = 1.
Q ( Y E , VEG) = Q (
I
I
11
REMARKS 9.2.8. (1) In view of Theorem 7.1.6 and Lemma 8.1.5, Theorem 9.2.1 holds not only for algebraic number fields K , but also for any field K and any n E N* such that gcd(n, e ( F ) ) = 1. Of course, the field C in the statement of Theorem 9.2.1 should be replaced by an algebraically closed field R containing K a s a subfield. Also, according to Theorem 7.2.3 and Lemma 8.1.5, Theorems 9.2.2 and 9.2.3 are valid for any field K , any n l , . . . , n , E N*, and any 21,. . . ,x, E R with x i k E K for all k, 1 k 6 r , such that gcd(n,e(K)) = 1 and p r L ( K ( x l ,.. . , x r ) ) E K , where n = lcm(nl , . . . ,n,).
Set E = Q( 'q26.33, and observe that E / Q is a 12-Kummer extension with few roots of unity. Then, by Theorem 7.3.2, we have
'I=),
'm, 'W).
where a = 24 . 3" b = 2' 3" and cl denotes for any x E p its coset XQ" in the quotient group Q+ /Q 12. Now we shall describe explicitly the group (2,;). Observe that a " (z4 . 36)6 E Q 12, hence words
<
(2) Corollary 9.2.5 and Theorem 9.2.6 can be extended frorn Q to the field of quotients of any UFD (see Exercises 1 arid 2). (3) We have seen that the conditions in Theorem 9.2.6, as well as that in Corollary 9.2.5 imply for K = Q the condition ( t t ) in Theorem 9.2.3. Observe that the extension Q(&, &)/Q satisfies the condition ( t t ) in x2 = &, but satisfies Theorem 9.2.3 with r = 2, nl = n2 = 2, xl = neither the conditions in Corollary 9.2.5 nor the conditions in Theorem 9.2.6. in Corollary 9.2.5, The extension Q(& & ) / Q satisfies the conditio~is but does not satisfy the conditions in Theorem 9.2.6. Also, the extension Q(&, m ) / Q satisfies the conditions in Theorem 9.2.6, but does not satisfy the ones in Corollary 9.2.5. This shows that we cannot deduce Corollary 9.2.5 frorn Theorem 9.2.6, and conversely.
-I:
i
g6 = 7, and ii# 7 for every i, 1 < i-< 6, in other '" , hence b 4 = 7, and gi # 7 ord ( 2 ) = 6. Also, b4 = (2" 33")" E
h h h
for every i , 1 hence
< i < 4, in other words ord ( b ) = 4. Moreover, b2 = a3 = 36,

h h h h h h h h h h h h h h h h h h h h h h h h h h h h h h h
h h
h h
h h
a,
h h
= { I , & a2, a3, a4, as, b, 2 . b, a" b, a 3 . b, a4s b, a"b)

h h
Note that i # ( g ) , and consequently I(;,;) I = 12. Thus [ E : Q ] = 12, and, by Corollary 2.1.10, a basis of the extension Q ( y E m ) / Q is
218
Chapter 9
9 . 3 Ilecke systems of ideal riurrlbers
9.3. Hecke systems of ideal numbers The Kneser Criterion (Theorem 2.2.1) has nice applications not only in investigating field extensions with Cogalois correspondence, but also in proving some very classical results from Algebraic Number Theory. In this section we present such applications. A classical construction from the Algebraic Number Theory is the following one: t o every algebraic number K one can associate a so-called system of ideal numbers S which is a subgroup of the multiplicative group C of complex numbers such that K * S and the quotient group S / K * is canonically isomorphic to the ideal class group CEK of K . This construction, originating with Hecke [70], has the following important property, that the Hilbert class field also possesses: every ideal of K becomes a principal 1 ideal in the algebraic number field K (S). The equality [ K (S) : K ] = leeK was claimed by Hecke, but never proved by him. The main aim of this section is to provide a short proof of this equality by using the Kneser Criterion, and t o discuss some other related questions.
has a unique decomposition where a E K * , 0 6 r k < hk, k = 1 , . . . , s , and the exponents r k are uniquely determined. Since cLk= 1, one deduces that I;* = (ck) E PK for suitable numbers ck E I<*, k = 1,. . . ,s , which are uniquely determined up to units of K . Assume that we have fixed the numbers ck and consider the number field K ( y l , . . . ,y,), where yk is a complex root of the polynomial X h k - ck E K[X], i.e., y ': = ck.
<
DEFINITION 9.3.1. W i t h the notation above, the group K * (yl , . . . ,r,) is called a Hecke system of ideal numbers of K , and the field K ( y l , . . . , y,), denoted by H K , is called the Hecke field of K associated with the Hecke 0 system of ideal numbers K * (yl , . . . ,y,) of K .
By Exercise 10, the Hecke field of any algebraic number field K , which clearly depends on the chosen Hecke system of ideal numbers of K , is uniquely determined up t o a K-isomorphism. Therefore, we will just call it the Hecke field of K .
/
I
First let us fix the notation and terminology needed to prove the equality mentioned above. Throughout this section K will denote a fixed algebraic number field. We will denote by D K the ring of algebraic integers of K , by 3 , ~ the group of nonzero fractional ideals of K , by PK the group of nonzero principal fractional ideals of K , by CEK = FK IFK the ideal elms group of K , and by h = ICEK] the class number of K . For any a E K * , (a) will denote the principal fractional ideal a D K of K . The following basic facts from the Algebraic Kumber Theory will be used several times in what follows without further comment. Let L / K be an extension of algebraic number fields. Then
I
I
I
,
#a; p,
The group rnorphism
FI( + K * ( y l,... ,y,)/K*, (a)IT1 . . . . .I,'.

with a E K * , 0 6 rnorphism
rk
H 7;'
.... eyy'.,
< Itk, k
= 1 , . . . S, clearly induces a surjective group
d K :FK/PK K * ( Y I , . . .,?'s)/K*LEMMA 9.3.2. With the notation above, the map

@K : C
I = I D L flDK = ID,, n K for any I E F K . If I , J E .FK are such that I ' " J7' for some n E N*, then I = J . As is well-known, the finite abelian group CEK is an internal direct sum of finitely many cyclic subgroups. This means that there exist ideal 1 , . . . ,C, in CCK, s 3 1, such that every ideal class C E CCK has classes C a unique decomposition c=crl . . . . . C ; s ,
a
~ K) K * ( y l , -- - ,?'s)/K*
is a group isomorphism. I n particular, one has

I K * ( y l , - . . ,?'s)/K*I = ICCKI = he
PROOF. We are going t o show that V K is injective. To do that, we have to prove that r~ = . . . = r, = O whenever y := 7:' . . . . y,Ta E K * , with 0 6 r k < hk, k = 1 , . . . ,s. Let us consider the integral ideal I := 1;'. .. *. .' ., I
Since
where 0 r k < hk, hk > 1 is the order of the ideal class Ck in CeK, k = 1 , . . . , s, and h = hl - . . . . 12,. For any k = 1 , . . . ,s let Ik be an integral ideal from the ideal class Ck. Then, every fractional ideal I E .FK
<
I : '
= (7k)'Lk= (ck),
220
Chapter 9
9 . 3 Ilecke systerns of ideal rlurrlbers
22 1
it follows that
PROOF.With the notation above, by Lemma 9.3.2, we have t o prove that [ K ( y 1 , . . . ,y,) : K ] = IK*(y1,... ,7s)/K*I = h. Since K ( y l , . . . ,7,) = K ( K *(yl ,. . . ,7,)) and yF1 = ci E K * for dl i = 1,.. . , s , it follows that the extension K(yl , . . . ,y,)/K is K * (yl , . . . ,7,)radical. Consequently, the desired equality means precisely that the extension K ( y l , . . . ,y,)/K is K * (yl , . . . ,y,)-Kneser. To prove that, we will check that the conditions from the Kneser Criterion are satisfied. Let p E P be an odd prime such that Cp E K * ( y l , .. . ,y,). Then Cp E K by Lemma 9.3.3. Now, assume that 1 C4 E K *(yl , . . . ,7,). Then (1 ~ 4 = ) 2C4 ~ E K * ( y l , . . . , y y ) , SO C4 E K * ( Y I , . . ,y,). Thus, C4 E K again by Lemma 9.3.3. Therefore, H KI K is a K * (yl , . . . ,?,)-Knew ex0 tension, which finishes the proof.
Ihk I~ ( I ~ D ~ jhk IK = ( Y ~ D I for all k = 1, . . . , s, hence

I k D l l ~= yk011~
for all k = 1,. . . ,s. Thus, ID,, and consequently
= (7;'
. . . . Y:)D~I~ = Y J ~ H ~ ,
I = I D H Kn D K = y o H K n D K = y o K .
This shows that the ideal class C of I in CeK is the identity class. Since
c=cyl .....c,.a,
we deduce' that r l = . . . = r , = 0, as desired. LEMMA 9.3.3. If e E K * (yl, . . . ,7,) is a u n i t of H K , t h e n
E
0
E K.
COROI,I,ARY 9.3.5. Let C be a n ideal class of K , and let y^ be the associated element in the group K *(y1 , . . . ,y,)/ K * via the isomorphism ?bK from L e m m a 9.3.2. If m = ord(C) in CCK, then [ K ( y ) : K ] = m , and every ideal of C becomes a principal ideal in K ( y ) .
PROOF. Since K*(y) 6 K * ( y I , .. . ,y,) and K ( y l , . .. ,y,)/K is a K * ( y l , . . . ,?,)-Knew extension by the proof of Theorem 9.3.4, it follows that K ( K *( y ) ) / K is a K *(7)-Kneser extension by Corollary 2.1.10. But K ( K * ( y ) ) = K ( y ) , hence [ K ( y ) : K ] = I K*(y)/K*I = Ord(T) = Ord(C) = na. By the proof of Lemma 9.3.2 we have IDH, = YDH, for any integral ideal I of the ideal class C. This implies that and we are done.
PROOF. Since e E K * ( y l , . . . ,ys), there exist a E K * , 0 6 r k < hk k = 1,. . . ,S, such that E = ayyl . . . . . y ? . From the proof of Lemma 9.3.2 we have I ~ D H K= Y ~ D H ~ for all k = 1 , . . . , s. Therefore, the fractional ideal I = (a)Irl . . . . Ii8 of K has the following property: IDH~ = (ayy - . . . - y?)Dl~, = We deduce that
= DfIK.
I = IDIIK n D K = DIIK n OK = DK = (1).

Consequently, the ideal class C of I in CeK is the identity class. Since
c=crl . . . . q
it follows that r l = . . . = r , = 0. Thus,
E
,
= a E K , and we are done.
THEOREM 9.3.4. Let K be a n algebraic number field, and let HK be its Hecke field. T h e n [HK: K ] = h.
R.EMARK 9.3.6. The Hecke field H K of a number field K has two of the basic properties of the Hilbert class field H C F K of K , namely: [HK : K ] = 12. a Every ideal of K becomes a principal ideal in H K . However, the fields H K and H C F K are different since the extension H C F K / K is always Galois, while, in general, the extension H K / K is not necessarily Galois. Corollary 9.3.5 shows that every ideal class C E CCK of order m becomes a principal ideal class in a suitable intermediate field
222
Chapter 9
9.3 llecke systerrls of ideal r~urr~bers
of H K / K of degree na over K . As is known, the question of whether H C F K has the same property, was answered in the negative by Artin and pp. 173-1741). FurtwPngler (see Hasse [68, 0
PROOF. (1) follows immediately from Proposition 9.3.7. (2) By ( I ) , we have [ K ( y j ) : K ] = h j for j = 1 , . . . , s, and so [ K ( y l,... ,y,) : K ] = h = hl . . . . ,Its = [ K ( y l ) : I < ] . ..: [K(ys) : K ] . This shows that the fields K (yl ), . . . ,K (7,) are linearly disjoint over K.
PROPOSITION 9.3.7. Let I E F K , let C the class of I in C ~ K let , m = ord(C) in CIK, and let c E K * with IT" = (c). Then, the polynomial X m - c is irreducible in K[X] .
PROOF. Let y denote a root in C of the polynomial X m - C. Then (ID K(,) )m = C ~ K ( , ) = (yDK(,))", hence I ~ K (= ,) yflK(,). Let E be a unit of OK(,) with E E K*(y). Then E = ayr for some a E K * and O 6 r < m. Hence
(3) follows from (2) and Lemma 8.1.5.
PROPOSITION 9.3.9. Let K be an algebraic number field, let SK = K * (yl, . . . ,y,) be any Hecke system of ideal numbers of K , and let H K = K*(yl , . . . ,y,) be its associated Hecke field. Then H K / K is an SK Kneser extension.
PROOF. The result follows a t once from the proof of Theorem 9.3.4. G The next example shows that, in general, the SK-Kneser extension H K / K is not necessarily SK-Cogdois.
Thus, I' is a principal fractional ideal, hence its ideal class Cr is the identity class. Since ortl(C) = m, it follows that nz 1 r, hence necessitrily r = 0, and so, E = a E K * . Now, proceed as in the proof of Theorem 9.3.4 t o deduce that K ( y ) / K is a K*(y)-Kneser extension. Therefore
E x . \ ~ r r9.3.10. .~ Let K = Q(J-87). It is known that the class number h of K is G (see e.g., the Table in Borevitch and Shafarevitch [37]). Since -87 1 (mod 4), it follows that D = [ L y ] . The decomposition of 3 as a product of prime ideals in the ring D K is
3DK = . 2)(3, 78- J If n = Ord(y^), then n (m, hence m = nt for some t E other hand, we have
W . On the
We claim that the prime ideal I = (3, of O K . Indeed, assume that I = ((6 for some
m )is not a principal ideal
hence 17"s a principal fractional ideal. As which implies that In= (r"), above, we deduce that m 1 n . Then m = n, and so j K (y) : K ] = m. We conclude that the polynomial X m - c is irreducible in K[X]. 0 COROI.~.ARY 9.3.8. With the notation above, the following statements hold. (1) The polynomials ~ " k ck are irreducible in K [ X ] for every k, l<k<s. (2) The fields K (7, ), . . . , K (7,) are linearly disjoint over K (3) There exists a canonical group isomorphism
and consider the norm N = N K I Q of the extension K l Q . Since (6) = (3, we deduce that N (6) I N (3) = 9 and N (6) 1 N = 87, hence
m),
(m)
N(6) = (a
\
i) +
2"
87
a ' + ab + 22b2 = 3.
.*:;
'1
This equation is not solvable in Z. Indeed, this is clear if ab 2 0. If ab < O then a%b2 > -ab and 21b2 2 21 > 3, hence weobtain a2+22b% 3-ab, and so, a' ab 22b" 3. Thus, the ideal class C1 of I in CtK has order 2, and I " (3) = (-3). as an ideal number of I . Since the group CeK is We choose y l = cyclic of order 6, it contains an ideal class C2 of order 3. Then, we have
+ +
221
Chapter 9
9.4 l?xercises to Chapter 9
225
Let y, be an ideal number of an integral ideal I 2 from the ideal class C2, that is, n , is one of the complex roots of the polynomial X 3 - e n E K[X], where cz E K * is such that 1; = (Q). Thus, the group K * (G , h)is a Hecke system of ideal numbers of K , and HK = K ( G , h)is its associated Hecke field of K . By Proposition 9.3.9, the extension H K / K is a K * ( m , 72)-Kneser extension. Observe h ) / K * ) = 6, 3 16 and = =%j@ E H K \ K, so the * that e x p ( ~(G, extension H K I K is not 6-pure. By the n-Purity Criterion, it follows that H K / K is not a K * ( G , T~)-Cogaloisextension. 0 The next result provides two cases when the extension H K / K is SKCogalois.

1. Let R be a UFD which is not a field, let K be its field of quotients, and let R be an algebraically closed field containing K as a subfield. Denote by PH a representative set of mutually nonassociated in divisibility nonzero prime elements of R, and by [ P R ] the subsemigroup of the multiplicative monoid (R, .) generated by pH, i.e., Let T E N, r >, 2, let n l , . . . ,n, E N*, let a l , . . . ,a, E [ P H I , and let X I , .. . ,x, E R be such that x : k = a k and o r d ( G ) = n k in R I K * for every k, 1 k r . Assume that gcd(n,e(K)) = 1, where n = lcm(nl, . . . ,n,), and p,&(K(xl,. . . ,x,)) C K . Then, prove that
PROPOSITION 9.3.11. The following statements hold for an algebraic number field K with class number h. (1) If G E K , then the extension K*(yl,. . . ,y,)/K is K * (yl , . . . ,7,)Cogalois for any choice of the Hecke system K *(71, . . . ,y,) of ideal numbers of I<. (2) If K can be embedded into the field R of real numbers, then there exists a Hecke system K *(71,. . . ,y,) of ideal numbers of K such that the extension K * ( y l , .. . ,y,)/K is K * ( y l , . . . ,y,)-Cogalois.
PROOF. (1) If Ch E K , then clearly p h ( K r ( y l, . . . ,y,)) C K . Since S , we deduce h = hl . . . . . h, and ykk = ck E K for all k, 1 $ k that K * (yl ,. . . ,y,)/K is a generalized h-Kummer extension. Kow apply Theorem 7.2.3 to obtain the desired result. ,
< <
[K("&
,... ,"G): K ] = n l ...:
n,,
provided the following condition is satisfied:
( )
a1 , . . . ,a k are relatively prime in pairs.
<
2. Show that Theorem 9.2.6 can be extended from Z t o a UFD satisfying the conditions in Exercise 1. 3. Show that [ Q ( l E , : Q ] = 8, and exhibit a vector space basis of the extension Q ( l S , a ) / Q .
:
(2) Without loss of generality, we may assume that K is a subfield of the field E3 of real numbers. In the construction of ideal numbers we may choose cl > 0,. . . , c , > 0. For yk we choose the positive real root of the polynomial X h k - ck, 1 k $ s. Then, HK = K ( y l , . . . ,y,) is a subfield of B. Thus, H K / K is a pure extension, and, a fortiori, an n-pure extension, where n = exp(Kt(yl , . . . ,y,)/K*). By the n-Purity Criterion, C1 we deduce that H K I K is a K *(yl , . . . ,7,)-Cogalois extension.
<
a) 4. Show that [ Q ( l % , m) Q ] = 18, and exhibit a vector space basis of the extension Q(q18, W ) / Q . 5. Find [Q(JZ, s, m) Q].
:
6. Let R be an algebraic closure of the field F := Q(X) of rational fractions in the indeterminate X , and consider the elements xl =
R.EMARK9.3.12. Let A be a Dedekind ring having a finite ideal class group C ~ A of order h , and denote by L its quotient field. Assume that the clwacteristic of L is not 2 and is relatively prime with h. Then, the rnain
part of the results presented in this section can be extended from algebraic number fields K to such more general fields L (see Exercises 11 and 12).
q m and
x2 =
yx4- 2x34-
2X2 - 2X
+1
of R. Find [ F ( x l , x 2 ): F ] .
7. Let
E N* and a ] , . . . ,a, E Z*. Prove that ,&)

:Q] = I V ( K , . . . ,&)lV
[Q(&,...
I.
226
Chapter 9
9.5 Bibliographical corrirrierits to Chapter 9
227
8. (Schinzel [93]). Let F be any field, let r E N*, let n l , . . . , n , E N* be positive integers a t most one of them divisible by the characteristic of F , let a l , . . . ,a, F * , and let X I , .. . , x, E R be such that xyi = ai for every i = 1, . . . r. Prove that
?
Section 9.1. The short review presented in this section is mainly based on R.ibenboim [89]. Section 9.2. The evaluation of the degree of a radical extension of type with K an algebraic number field containing C,, K ( m , . . . , *)/K n, r E W , and a l , . . . ,a, E K * , as the order of the quotient group K *( , . . . , ;/a,) / K * has been known for many years by the classical Kummer Theory. The problem "When is [ K ( m , .. . , :K j = iiir [ K ( G ) : K I?" was first answered by Hasse in 1930 (see the second edition [68] of his mimeographed lectures on Class Field Theory ). The case when the algebraic number field K does not necessarily contain a primitive n-th root Cn of unity was, surprisingly, first discussed fairly late, only in 1940 by Besicovitch [35] for K = Q and . . . , fi real roots of positive integers a1 , . . .a,, and then, by Mordell [83] in 1953 for K any real number field and m , . . . , i/ar E R. Later, in 1972, Siege1 [98] shows that the degree [ K ( m , . . . , : K ] is the order of the quotient group K * ( . . . ,&) / K * for any real number field K and any real roots ,. . . ,*. A particular case of Corollary 9.2.7 was proved by Richards [go] (see also Gaal's book [58], where Richards' proof is reproduced). Corollary 9.2.5 is due to Crsell [106], but our proof is different from his. The presentation of this section is based on Schinzel [93] and Albu [3].
if and only if the following two conditions are satisfied. (a) Whenever p E P and k l , . . . ,k, E N are such that p 1 niki : . E FP, then p 1 ki for for every i = 1,. . . ,r and a:'. . . . . a every i = I , . . . ,r. (b) Whenever kl, . . . ,k, E N are such that 4 1 niki for every i = 1 , . . . ,r and a:' . . . . a:? E -4F4, then p 1 ki for every i = l , ... ,T. With the notation of Section 9.3, show that for every a E K* one has ayT1.. .. - y?DIIK n K = (a)Ilrl .. - Isr8.
a .
n,
s)
m,
Prove that the Hecke field of any algebraic number field K is uniquely determined up to a K-isomorphism. More precisely, with the notation of Section 9.3, show that if y; is another root of the polynomials Xhk - ck, k = 1,. . . ,s , then the fields K ( y l , . . . ,7,) and K ( y i , . . . ,yL) are K-isomorphic via a K-isomorphism sending yk to yk, k = l , ... , s . Let I? be a Dedekind ring having a finite ideal class group C I A of order h, and denote by L its field of quotients. Assume that the characteristic of L is not 2 and is relatively prime with h. Show that we can perforrn mutatis-mutandis the construction presented in Section 9.3 to define a Hecke system L* (yl, . . . ,7,) of ideal elements of L and a Hecke field HI, of L. Prove that if R is a Dedekind domain as in Exercise 11, and L is its field of quotients, then Theorem 9.3.4, Corollary 9.3.5, Proposition 9.3.7 and Corollary 9.3.8 also hold for the field L. Let K1 and K 2 be algebraic number fields such that their discriminants S K , and dK, are relatively prime. Prove that [ K 1 K 2: Q ] = [ K , : Q ] . [ K 2 : Q].
m,
s)
t \I
f #
Section 9.3. Hecke systems of ideal numbers K *( y l , . . . ,y,) associated with any algebraic number field K were introduced in 1920 by Hecke [70]. In his rnonograph [71] published in 1948, Hecke claims on page 122 that the degree [ H K : K ] is precisely the class number h of K , where H K is the field K ( y l , . . . ,y,) associated with the systern K * ( T I ,. . . -yY) of ideal numbers. To the best of our knowledge, no proof of this assertion excepting the one in Albu and Kicolae [21]is available in the literature. Note that Ribenboim gives on page 124 of his monograph [89] only the inequality [ H K :K ] h. Related t o Hecke systems of ideal numbers a natural question arose: are the polynotnials x h k - ck irreducible in K[X], where ck = y i k E K? This problem was only mentioned (but not settled) by Hasse [67, p. 5441 as
<
228
Chapter 9
K wollen wir hier nicht eingehen." The positive answer to this question was given by Albu and Nicolae [21]. The results of this section are taken from Albu and Nicolae [21].
CHAPTER 10
CONNECTIONS WITH GRADED ALGEBRAS AND HOPF ALGEBRAS

The aim of this chapter is t o present some interesting connections of Cogalois Theory with graded algebras and Hopf algebras. In Section 10.1 we analyze the concepts of G-radical, G-Kneser, and G-Cogalois extension in terms of graded algebras. After reviewing some basic notions and facts on Hopf algebras, we describe in Section 10.2 the G-Kneser extensions and Cogalois extensions in terms of Galois H-objects appearing in Hopf algebras.
10.1. G-Cogalois extensions via strongly g r a d e d fields
In this section we describe the concepts of G-radical, G-Kneser, and G-Cogalois extension in terms of graded ring theory. Throughout this section all algebras are assumed to be associative with unit, and K will denote a fixed commutative ring with nonzero identity element. If A is a K-algebra and X, Y are subsets of A, then XY will denote the K-submodule of the underlying K-module of the algebra A which is generated by the set { xy 1 x E X, y E Y ). Any ring can be viewed in a canonical way as a Z-algebra. For a K-module M and a family (Mi)iEl of submodules of M , the Mi will mean throughout this chapter that M is the notation M = "internal direct sum" of the independent family (Mi)iEl of its submodules, Xi, that is, any element x E M can be uniquely expressed as x = CiEI where is a family of finite support, with xi E Mi for every i E I.
eiEl
DEFINITION 10.1.1. Let I? be a multiplicative group with identity elem e n t e . A K-algebra A is said o be a I?-graded algebra if A = A,
230
Chapter 10
10.1
G-Cogalois exterlsiorls via
strongly graded fields
23 1
is a direct s u m of K-submodules A, of A, with A,Aa S Ay6 for every y, 6 E r. A r-graded algebra A = A, is said to be strongly graded if A,A6 = AyS for every y, 6 E r . A (strongly) r-graded ring is a (strongly) r-graded algebra over the ring Z of rational integers. A, is said to A left module A1 over the r-graded algebra A = be a graded module (resp. a strongly graded module) if M = My is a direct s u m of K-submodules MT of M , with A,M6 E My6 (resp. .4,M6 = MTs) for every y, 6 E r . The elements of h ( M ) = UTErM, are called the homogeneous elements of M . A n y element x E M has a unique x,, with 2, E M,, 7 E r , and all but a finite decomposition x = CTtr number of the x, are zero; the elements x, are called the homogeneous 0 components of x.
eTEr
Conversely, if there exists a family ( x , ) , ~ ~ ,E E S which is a basis of the vector space r E , then EIF is a strongly r-graded extension. ( 4 ) If E/F is a finite extension, then E/F is a strongly r-graded extension if and only if
nTEr
(5) The group r is Abelian. (6) If r is a torsion group, then E/F is a n algebraic extension.
PROOF. (1) If E, = E 6 , then
by Definition 10.1.2 (2), hence 67-l = e, i.e., (3). E,-1
-y =
DEFINITION 10.1.2. Let EIF be a field extension, and let I' be a multiplicative group with identity element e. One says that E/F is a r-Clifford extension (resp. a strongly r-graded extension) if there exists a family (Er)yEl. of F-subspaces of 'FE indexed by the group I?, satisfying the following conditions.
6 by Definition 10.1.2
,
(2) Again by Definition 10.1.2 (2), we have E,,-I = E, = F , hence # 0. Ta,ke an arbitrary y E E;-, . Then, the map
E,
+ F,
I-+
x y , x E E,
yields an isomorphism of F-vector spaces. (3) By (2), E, = Fx, for every y E r , hence
A n element x E E is said to be homogeneous of degree y if x E E,. The set of all nonzero homogeneous elements of E will be denoted by uh (E). 0
Note that we do not assume any finiteness condition on a r-Clifford extension, though we are mainly interested in finite extensions.
E=
,I'
E, =
Fx,,
yEI '
LEMMA 10.1.3. The following statements hold for a I?-Clifford extension E/ F.

(1) If y, 6 E r, then E, = E d e y = 6. (2) dimr(E,) = 1 for all y E T . I n particular, E, = FI: for every x E E,:. (3) For every y E I? let x, E E ; be arbitrary. Then X = { x, 1 y E r ) is a set of generators of the vector space v E , and [ E : F ] 6 I I' 1. If E/F is a strongly r-graded extension, then X is a basis of the vector space F E , and [ E : F] =
i.e., X is a set of generators of the vector space F E . This clearly implies that [E:FI<IXl<lrl. By definition, the r-Clifford extension E/F is strongly I?-graded if and only if
i.e., if and only if the set of generators X of the vector space v E is linearly independent over F , hence a basis of ,G.E,and in this case we clearly have [ E : F] = ( X I. Now, observe that if E/F is a strongly r-graded extension, then the map r -+ X, y I+ x,, is bijective, hence 1 I' 1 = I X I. of (4) follows from (3) since in any finite dimensional vector space dimension n, any set of generators S of FV with IS1 = n is necessarily a basis of F V .
lrl.
232
Chapter 10
10.1 G-Cogalois exterisior~s via strongly graded fields
233
(5) Since the multiplication on E is commutative, we have hence 76 = Sy for every y,S E
by (1).
(6) Since I? is torsion group, it follows that for every y E r and every x, E E, there exists an n, E N* such that 7". = e, hence x;' E E ; . = Ern, = E, = F . Thus, every homogeneous element of E is algebraic over F, and since any element of E is a sum of homogeneous elements, it follows that E/F is an algebraic extension. 0 The next example shows that, in general, the converse implication in Lernma 10.1.3 (6) does not hold.
- .
E/Jt- 1 , hence that 7 = X1 . . XrP;l . . p;l. But E7 = EAl . - Ex,,EP;l we deduce that E ( A ) K. For the opposite inclusion, let y E K. Since E/F is an algebraic extension, we have K = F ( M ) = F[M], hence there exists a finite subset {A1,. . . ,An) of A such that y can be written as a finite sum of mono. . . xkn, with a E F and i l , . . . ,in E N. Since a r t l . . E mials E ( A ) , it follows that y E E(A), which proves the inclusion E,;,
C E(A).
P R O P O ~ ~ T I O 10.1.6. N
.,,
s?'
0
The following assertions hold for a field exten-
sion E I F . G E * . If E/F is a G-radical Let G be a group with F* extension, then E/ F is a G I F *-Clifford extension. Conversely, if E/F is a r-Clifford extension for some torsion group E*, r r r , then there exists a group G such that F* G G / F * , E/F is G-radical, and U h ( E ) = G. If E/F is a G I F *-Cliflord extension, where F* G 6 T ( E / F ) and Uh ( E ) E G, then E/F is G-radical. Let G be a group with F* G E * . If E/F is a finite G-Kneser extension, then E/F is a strongly G I F *-graded extension. Conversely, if E/F is a strongly r-graded ez$ension for some finite group r , then there exists a group G such that F* G E*, I? 2 G / F * , E/F is G-Kneser, and Uh(E) = G. If E/F is a strongly G I F *-graded extension, with F* G E*, G I F * finite and U h ( E ) C G, then E/F is G-Kneser.
<
<
E/F is a quartic extension, hence an algebraic extension, which is also a Z-Clifford extension with
E = X E , , , E,, = FB", n E Z .
Indeed, the only nontrivial fact is that En = F + n = 0. But En = F clearly implies that 0" E Q, which can happen only if n = 0 by Proposition 3.2.6 (g). Thus, E/F is a %Clifford extension, but Z is a torsion-free group. 0 For any r-Clifford extension E/F and any A
< <
<
< <
< I? we shall denote
< < < <
Obviously, E d F is a A-Clifford extension, which is strongly A-graded whenever the extension E/F is strongly I?-graded. LEMMA 10.1.5. Let E/F be a I?-Clifford extension, let 0 # A E I?, and let xx E El;, X E A. If E/F is an algebraic extension (in particular, if r is a torsion group), then
PROOF. (1) Assume that E/F is a G-radical extension. Then, by Remark 2.1.4 (3), we have E = F ( G ) = F[G], hence G generates E as a vector space over F , i.e.,
h
I-
E ( A= ) F({xA
A}).
E G, then Fgl = F g L +=+ 97 = 6, hence it makes Observe that if g l , g . ~ sense to set &j= F g for every g E G. Then clearly
PROOF.Set M = {xx I X E A} and K = F ( M ) . Since dimF(EA)= 1 and X A E Ex n M , it follows that E x K for every X E A. Further, if X E A and x E E f - , , then a = xxx E EX-1Ex= Ee = F , hence x = ax;' E K. This shows that Ex-1 E K. Now, let y E (A). Since r is Abelian by Lemma 10.1.3 (5), there exist XI, . . . , A,, p l , . . . ,pt E A such
234
Chapter 10
10.1 G-Cogalois exterlsior~s via strongly graded fields
235
and
E,?E & = F91 F g t 2 = 0 1 9" = E9z2= Eli; &, , which show that E/F is a G/F*-Clifford extension.
(2) Assume that E/F is a I?-Clifford extension with E = ET. Pick for every y E I? an arbitrary element x, E E;. Then E, = Fx, by Lemma 10.1.3 (2). For any y, 6 E I? we have X,xa E E, E6 = E,a, so there exists a,,6 E F* such that x,xa = a,,ax,a. This shows that the map
= ? implies that is a group rnorphisrn. This morphism is injective since x, E F * , hence E, = c x , = F , and so, y = e by Definition 10.1.2 (3). Set G = Im(w). Then G Z r , and there exists G with F* G E* and G = G I F * . Observe that
On the other hand, since E / F is G-Kneser, S is a vector space basis of E over F by Proposition 2.1.8. Therefore
which shows that E/F is a strongly G/F*-graded extension. (5) According to (2), there exists a group G such that F* G E * , I? 2 G I F * , EIF is G-radical, and U h ( E ) = G. Pick for every y E I? an arbitrary element x, E E;. Then, the set B = { x, 1 y E I? ) is a basis of the vector space pE by Lemrna 10.1.3 (3), and clearly B Ufl(E) = G. By the proof of (2), g # for every y # S in I?, so B is a set of 1y E I '). Now, apply representatives of the group G = G / F * = {g Corollary 2.1.10, to deduce that E/F is a G-Kneser extension.
(6) Since EIF is a strongly G/F*-graded extension, we have E = E?, hence [ E : F] = ( G IF*( since dimp(Ez) = 1 for all $ E
< <
< <
hence E=E~=F({X,J~EI'))=F(G) by Lemma 10.1.5. We have F*x, = F* 6 G. Since is a torsion group, it follows that G T ( E / F ) , hence E/F is a G-radical extension.
G I F * by Lemma 10.1.3 (2). By (3), the extension EIF is G-radical, 0 hence it is G-Kneser. The graded version of the concept of G-Cogalois extension is that of I?-Cogalois extension we are going t o introduce below. E,, and any interFor any I?-Clifford extension EIF with E = CrEr mediate field K of E/F we shall use the following notation: I'K={~EI?~E,~K). Then clearly map
<
(3) For every g^ E G / F * pick an x i E E;. Since E;=u"(E)~G $EG/F* by hypothesis, we have { x i 1 g^ E G I F * ) 5 G. Now, applying Lemma 10.1.5, with A = I? = G I F * , we deduce that
rKis a subgroup of
I ' ,hence it makes sense t o consider the

@(K) =r ~ ,
@ : Intermediate(E1F)
+ Subgroups(T),
E = E c / p = F({X? ( 5E G I F * )) E F ( G ) E E ,
i.e., E = F ( G ) . This shows that EIF is a G-radical extension. (4) By ( I ) , EIF is a G/F*-Clifford extension, with
which is clearly order-preserving. Also, we have an order-preserving map
Clearly, the maps 9 and @ define a Cogalois connection between the lattices Intermediate(E/F) and Subgroup~(I?). This implies that if S is any fixed set of representatives of G I F * , then
LEMMA 10.1.7. With the notation above, the following statements hold for any I?-Clifford extension E / F . (1) E / K is a I'/rK-C1ifford extension for every intermediate field K of E I F .
236
Chapter 10
10.1 G-Cogalois exterlsions via strongly graded fields
(2) If E/F is a strongly r-graded extension, then @o P = lSubgroups~l~~. PROOF. (1) For any y E I' we denote by y^ its coset modulo the subgroup r K of I". Let yl, 7 2 E I", with y^l = $. Then y . = ~ y16 for some 6 E r K . Thus E a K , and so E a K = K . It follows that
(3) E/F is a strongly I?-graded extension, and the maps
Q, : Intermediate(E1F) ---+ Subgroups(I"), @(K)= I"K, B : Subgroups(I') + Intermediate(E/F), #(A) = EA

are isomorphisms of lattices, inverse to one another. (4) E/F is a strongly I?-graded extension, and every intermediate field K of E/F has a vector space basis over F consisting of homogeneous elements. (5) EIF is a strongly I'-graded extension, and every intermediate field K of E/F is obtained by adjoining to F a set of homogeneous elements of E. (6) EIF is a strongly I"-graded extension, and every nonzero element x E E has its homogeneous components in F ( x ) . ( 7 ) E / K is a strongly I"/I"K-graded extension for every intermediate field K of E / F .
PROOF. (1) 3 (2): Taking K = E , we obtain from hypothesis that E / F is a strongly rri-graded extension, hence a strongly I'-graded exten= I". Frorn Lemma 10.1.7 (2) we deduce that @ sion, since obviously is surjective. E, for every intermediate field K of We claim that K = E / F . By hypothesis, K / F is a strongly I'K-graded extension, hence K = K,. Since E = E,, we clearly have K, E, for every K,. y E r K . Conversely, for any y E I"K we have E, C K , hence E, This proves our claim. We are going to show that @ is injective. Let K1, K 2 be two intermediate fields of EIF such that @(K1)= Q,(K2),ix., = rK2. Then K1 = $ E,= $ E , = K 2 .
Therefore, it makes sense to set E9 = E,K for every y E I ' . For any we have 7 , 71, 7 2 E
ECEG = (E,, K)(E,,K) = E,, E,,K
= E,,,,K
= EFG.
Let S be a set of representatives of the quotient group {~I"K I a E S ) is a partition of r, hence
r/rK.Then
= ~ ( E , K )=
u ES
q.
%~/I'K This proves that E I K is a I'/rK-Clifford extension.
I ' we have (Q,0 #)(A) = rEA = A. Indeed, the (2) For any A inclusion A C I"E, is obvious. For the opposite inclusion, let y E r E A . EA = C,,, E6, hence necessarily y E A since the assumpThen E, EA= tion that E/F is a st;6ngly I"-graded extension means that CAEr @xcr Ex.
<
rKl
YE~K,
~EI'K,
(2) w (3) follows immediately from Lemma 10.1.7 (2).
PROPOSITION 10.1.8. The following assertions are equivalent for a I"Cliflord extension E/F with I' a torsion group. (1) Every subextension K / F of EIF is a strongly rrc -graded extension. (2) E/F is a strongly r-graded extension, and the map
(3) ==+ (4): Let K E Intermediate(E/F). Then K = ( P o @)(K) = E-,. This shows that K , viewed as an F-vector space has a El., = basis consisting of homogeneous elements {x, I y E rK), where x, is an arbitrary nonzero element of E, for every y E r K .
(4) =3 (5) is obvious.
is a lattice isomorphism.
(5) (6) By Lemma 10.1.3 (6), the extension E/F is algebraic. Let x E E * . By assumption, there exists a finite subset {yl , . . . ,y,)
238
Chapter 10
10.1 G-Cogalois exterlsions via strongly graded fields
239
of I' and x , E E;,, 1 i n, such that F ( x ) = F ( x,,,. .. ,x,,,) = F[x,, , . . . ,x,.~]. Thus, every element of F ( x ) can be written as a finite sum of monomials ax;, . . x k , with a E F and i l , . . . ,in E N. This implies that F ( x ) = Ea, where A = ( y , , . . . ,y , , ) . But E/F is a strongly r-graded extension, hence every homogeneous component of x belongs t o EA,i.e., to F ( x ) . (7): Let K / F be a subextension of E / F . By Lemma 10.1.3 (6), 0 # A K . Let C be the set of all nonzero homogeneous components of all elements of A. By assumption, we have K = F[C]. Let D = { y E I? 13 c E C, c E E, ) and set A = (D). Then, as in the proof of implication (5) (6), it follows that K = Ea = CaEa Ea = E6,that is, K / F is a strongly 4-graded extension. Fro~n Lem~na 10.1.7 (2) we deduce that A = (@ 0 #)(A) = rEA =rK. According to Lemma 10.1.7 ( I ) , E / K is a I'll-'K-Clifford extension, with (6)
< <
Let K be an intermediate field of the extension E / F . Then E / K is a strongly r/rK-graded extension by hypothesis. Hence
where S is a fixed set of representatives of the quotient group I'/I?K. As (7), we have in the proof of implication (6)
EIF is an algebraic extension, so K = F[A] for some
edEA
since clearly
XE~K
Using the definition of the internal direct sum of submodules of a module, we deduce that
\XEI'~
'
Let S be a set of representatives of the quotient group r / r K . Since {~ K IU u E S ) is a partition of I?, then as in the proof of Lemma 10.1.7 (1) we have
for every a E S. By Definition 10.1.2, this implies that, for a fixed a E S we have
which proves that the extension K I F is strongly I?-graded.
DEFINITION 10.1.9. A field extension EIF is said to be I?-CliffordCogalois if it a finite separable I'-Clifford extension which satisfies one of 0 the equivalent conditions of Proposition 10.1.8.
which shows exactly that E / K is a strongly I'll?K-graded extension. (7) ==+ (1) If y E rp, then E, C F, hence 1 = dimr.(E,) dim~(F= ) 1. Tlms E, = F , and then y = e by Definition 10.1.2 (3). This shows that r ~ =. {e). Now, taking K = F in (7), we deduce tha.t E/F is a strongly r-graded extension, hence E = $?,,. E,.
<
PROPOSITION 10.1.10. The following statements hold for a finite field extension E/ F . (1) If the extension E/F is G-Cogalois for some G with F* G E * , then E/ F is a GI F*- Clifford-Cogalois extension. (2) Conversely, if E/F is a r-Clifford-Cogalois extension for some finite group I?, then there exists a uniquely determined group G such that F* < G < E*, I? Z G / F * , E/F is G-Cogalois, and Uh(E) = G.
< <
240
Chapter 10
10.1 G-Cogalois extensions via stror~glygraded fields
PROOF. (1) By Proposition 10.1.6 (4), E/F is a strongly G/F*-graded extension, and by Theorem 4.2.7, the maps
with the lattice isomorphism
9 : Subgroups(r)
-+ Intermediate(E/F),
@(A) = E A ,
given by Proposition 10.1.8, we obtain the lattice isomorphism are isomorphisms of lattices, inverse t o one another, where But El.,.,,,, = F ( H ) by Proposition 10.1.8, hence @ o r is exactly the map
G = { H I F*< H
< G ) and
& = Intermediate(E/F).
Clearly, the canonical map p : Subgroups(G/F*) -+ G, H/F* I+ H , is a lattice isomorphism, so the composed map is a lattice isomorphism. By the proof of Proposition 10.1.6 (1) we have
1C, : G --+ &, $(H) = F(H). Now, apply Theorem 4.2.7 t o conclude that E/F is G-Cogalois. uniqueness of G follows from Corollary 4.4.2.
The
17
The next result is the graded version of the uniqueness of the Kneser group of a G-Cogalois extension (see Corollary 4.4.2). COROLLARY 10.1.11. Let E/F be a finite extension which is simulta' neously I? -Clifford- Cogalois and A -Clifford- Cogalois. Then the groups I and A are isomorphic.
E H J p=
so
F h = F[H] = F(H),
~IIJF*
PROOF. Apply Proposition 10.1.10 and Corollary 4.4.2.
( d P) ( H I F * = F ( H ) = EHJ F* for every H E G. This shows that $ o p is precisely the canonical map P : Subgroups(G/F*) -+ Intermediate(E/F), P ( H / F * ) =
PROPOSITION 10.1.12. Let E/F be a r-Clifford-Cogalois, let A be a xx finite nonempty subset of I ' , and let { xx I X E A) Uh(E). Then CAE,, is a primitive element of the extension EIF if and only if ( A ) = r.
PROOF.Denote K = F(CxE,, zx). By Proposition 10.1.8, we have xx E K for every X E A, hence F ( ( x x I X E A )) C K. The other inclusion K E F ({ zx ( X E A )) is clear, hence K = F ({ xx I X E A 1). On the other hand, E(A ) = F({xx I X E A ) ) by Lemma 10.1.5. Thus, by Proposition 10.1.8, we have
defined just above Lemma 10.1.7. Thus, the canonical map @ in Proposition 10.1.8 is a lattice isomorphism by Lemma 10.1.7 (2). By Proposition 10.1.8, it follows that E/F is a G/F*-Clifford-Cogalois extension. (2) According t o Proposition 10.1.6 (5), there exists a group G such that F* G E * , I? 2 G I F * , E/F is G-Kneser, and u h ( E ) = G. Preserve the notation from the proof of Proposition 10.1.6 (2). The group isomorphism w : I' -+ G / F * induces the lattice isomorphism
< <
T : G -+ Subgroups(I'), HI+ { y E I'lE; We claim that for any y E r ,
H).
One implication is clear. For the other one, if E, C F ( H ) , then we have F ( H ) n G = H by Proposition 2.1.11. Therefore, T ( H ) = I ' F ( H ) . E ; Consequently, if we compose the lattice isomorphism
REMARK 10.1.13. The characterization of finite G-Kneser and G-Cogalois extensions in terms of graded algebras given in Proposition 10.1.6 (4)(6) and Proposition 10.1.10, respectively, also hold for infinite extensions. This can be easily seen if in their proofs one simply invokes Proposition 11.1.2 and Theorem 11.2.6 instead of Corollary 2.1.10 and Theorem 4.2.7, respectively. See also Exercise 5 in Chapter 11 and Exercise 14 in Chapter 12. Note that Corollary 10.1.11 has also an infinite variant (see Exercise 0 15, Chapter 12).
242
Chapter 10
10.2 Cogalois exterlsions and I-lopf algebras
243
10.2. Cogalois extensions and Hopf algebras In this section we describe the property of a finite field extension being G-Kneser, in particular Cogalois, in terms of Hopf algebras. To do that, we present first the main concepts and facts on Hopf algebras which will be used in that description. Throughout this section K will denote a fixed field, and H will denote a K-Hopf algebra. Tensor products are assumed to be over K , unless stated otherwise. We shall denote by K-Alg the category of all associative K algebras with unit, and by K-Mod the category of all K-vector spaces. If A E K-Alg, then Mod-A will denote the category of all unital right A-modules. 10.2.1. ALGEBRA AND COALGERRA. We first present a n alternative definition for the classical concept of an associative K-algebra with unit via maps and diagrams, so that we may dualize it. A K-algebra is a triple (A,p, u), where A is a K-vector space, p : A @ A A A and u : K + A are morphisms of K-vector spaces, called the multiplication map and the unit map of A, respectively, such that the following diagrams are commutative:
We denoted by l Athe identity map of A, and the unnamed arrows from the second diagram are the canonical isomorphisms. We now dualize the notion of algebra. A K-coalgebra is a triple ( C , A , z ) , where C is a K-vector space, A : C -+ C @ C and z : C + K are morphisms of K-vector spaces, called the comultiplication map and the counit map of C , respectively, such that the following diagrams are commutative:
Let (C, A , E ) be a K-coalgebra and let c E C. Then
i n. The sigma notation, for some n E N* and cil, cia E C, 1 introduced by Heyneman and Sweedler, suppresses the subscript i, so that,
< <
tI
If (A, /LA, u A ) and (B, p11, uU) are two K-algebras, then observe that a K-linear map f : A -+ B is an algebra morphism if and only if the
Chapter 10
10.2 Cogalois extensions and Hopf algebras
245
This observation allows us t o define the dual concept of coalgebra morbe two K-coalgebras. phism as follows. Let (C, Ac, ec) and (D, AD, Then, a K-linear map g : C -+ D is a coalgebra morphism if the following diagrams are commutative:
Similarly, one can define right modules over the algebra A, the only difference being that the structure map of the right A-module X has the form v : X @ A -+ X. If X is a left or right A-module, one also says that A acts on X . By dualization, we obtain the concept of a comodule over a coalgebra. Let (C, A, E ) be a K-coalgebra. A right C-comodule is a pair (M, p), where M is a K-vector space and p : M --+ M @ C is a morphism of K-vector spaces, such that the following diagrams are commutative:
The commutativity of the first diagram may be written in the sigma notation as: A D ( ~ c )= ) g(c)l @ ~ ( ~ = 1 2 g(c1) @ 9 ( h ) .
If M is a right C-comodule, one also says that C coacts on M.

10.2.2. COMODULE. In the same way as in the previous subsection, where we gave an alternative definition for an algebra using only morphisms and diagrams, we begin this subsection by presenting an alternative definition of the concept of left module over an algebra. Let (A, p, u) be a K-algebra. A left A-module is a pair (X, v), where X is a K-vector space and u : A @ I X --+ X is a rnorphism of K-vector spaces, such that the following diagrams are commutative:
We are now going to define the morphisms of cornodules. To do that, we present first the definition of a module morphism using only commutative diagrams. Let A be a K-algebra, and let (X, vx), (Y, u Y ) be two left A-modules. The K-linear map f : X -+ Y is a morphism of A-modules if the following diagram is commutative:
Chapter 10
10.2 Cogalois exter~siorlsand Ilopf algebras
247
10.2.3. B I A I , G E R R A A N D HOPFALGEBRA. A K-bialgebra is a K-vector space B endowed with an algebra structure ( B , p , u ) and with a coalgebra structure (B,A,c^), such that A and E are algebra morphisms, or equivalently, p and u are coalgebra morphisms. A map f : B --+ B' of bialgebras is called a bialgebra morphism if it is both an algebra morphism and a coalgebra morphism.
S :H
By dualization, we define the concept of a comodule morphe'sm. Let C be a K-coalgebra, and let (M, p ~ ) (, N , p N ) be two right C-comodules. The K-linear map g : M -+ N is called a morphism of C-comodules if the following diagram is commutative:
A , E ) having a K-linear map A K-Hopf algebra is a bialgebra (H, p , u, + H satisfying the condition
for every h E H , where 1 is the unit element of H . The map S is called a.11antipode for H . -4map f : H -+ H' between two K-Hopf algebras H and H' is called a Hopf morphism if it is a morphisrn of bialgebras. Any Hopf morphism preserves antipodes.
DEFINITION 10.2.4. A left H-module algebra is a K-algebra A such that A is a left H-module with structure 7nap
H@A-+A, h@a~+h.a, ~ E H~ , E A , and such that the multiplication p : A 8 A + A, p ( a @ b) = ab is a morphism of H-modules, where A dD A is a left H-module via he ( a @ b) = E ( h l - a ) @ (It2. b), h E H , a , b E A. For any right C-comodule M with structure map p : M and any m E M , we shall also use the sigma notation:
-+ M QD C
0
DEFINITION 10.2.5. A right H-cornodule algebra is a K-algebra A such that A is a right H-comodule with structure map
p:A-+A@H, p(a)=Eao8al, ~ E A , and p is an algebra n~orphism.
Thus, the commutativity of the diagram above involving the definition of a comodule morphism, may be written in the sigma notation as:
DEFIXITION 10.2.6. A Galois H-object is a right H-comodule algebra A such that the linear map
y : A @ A - - + A @ H , ? ( a @ b ) = x a b u ~ b la, , b ~ A ,
For a K-coalgebra C we shall denote by Comod-C the category of all right C-comodules.
is bijective.
DEFINITION 10.2.7. Let A be a right H-comodule algebra. A right (H, A)-Hopf module, or a Doi-Hopf module is a K-vector space M such that the following three conditions are satisfied.
248
Chapter 10
10.2 Cogalois exter~sior~s and Ilopf algebras
249
(1) M is a right H-comodule via the map
(2) The functors F and G in Proposition 10.2.10 are inverse equiualences, or equivalently, for all X E K-Mod and all M E M f, the natural maps
(2) M is a right A-module via the map
are isomorphisms. In the sequel, Mf will denote the category of all right (H, A)-Hopf modules, where the morphisms are the maps which are simultaneously morphisms of right A-modules and morphisms of right H-comodules. EXAMPLES 10.2.8. (1) Clearly, A E MY, with the usual structure of right A-module of A, and with the map p : A + A @ H arising from the structure of right H-comodule algebra of A. (2) If X E K-Mod, then A @ X E M i , with action (a @ x) . b = ab 8 x, and coaction ~) a l , a 8 x ct x ( a 0 C O X for a , b E A, x E X . DEFINITION 10.2.9. Let M E Comod-H with comodule structure given by the map v : M + M 8 H . The set is a K-vector subspace of M which is called the subspace of H-coinvariants of M . In particular, for any right H-comodule algebra A we can consider A''", which is even a subalgebra of A. 0 is an equivalence of categories. The corollary now follows from a general result saying that whenever we have an equivalence Moreover, if (1) or (2) holds, then ACoH= K . PROOF. See Caenepeel [42, Definition 8.1.5 and Theorem 8.1.61.
COROLLARY 10.2.12. Let A be a Galois H-object. Then, for any M E
M:, the canonical map

is a lattice isomorphism. PROOF. By Theorem 10.2.11, the functor
between two arbitrary categories C and V, then for every X E C, the functor G induces a bijective map between the subobjects of X and the subobjects of G(X). 0 Kext, we will illustrate the concepts of right H-comodule algebra and Galois H-object with a basic example. Let G be any multiplicative group with identity element e, and let H = K[G] be the group algebra of G over the field K. It is well-known that K[G] is a Hopf algebra over K via the ) 1, and cornultiplication given by A(g) = g @ g, the counit given by ~ ( g = the antipode given by S(g) = g-' , for all g E G. LEMMA 10.2.13. Let G be a group, and let M be a K-vector space. Assume that there exists a family (M!])!]G of K-subspaces of M such that M = CpEG MR. Then, p : M + M @K K[G], x u x 8 g, for all x E A , ] ! and in this case M is is a well-defined map if and only if M = a right K[G] -comodule via the linear map p.
PROPOSITION 10.2.10. Let A be an H-comodule algebra, and consider the functors:

F:K-Mod-+M14/, G: F(X)=A@X,XEK-Mod, G ( M ) = M'O",
MY
-+ K-Mod,
M E MY.
Then, G is a right adjoint of F .

PROOF. See Caenepeel [42, Proposition 8.1.31.
0
$
THEOREM 10.2.11. The following statements are equivalent for a Hopf algebra H with bijective antipode and for a right H-comodule algebra A. (1) A is a Galois H-object.
260
Chapter 10
10.2 Cogalois extensior~s arid Ilopf algebras
26 1
Conversely, if the K-vector space M is a right K[G]-comodule,then M is a G-graded K-vector space, that is, M = N, is a direct sum of K-subspaces N,, g E G.
eIEG
PROOF. Assume that p is a well-defined map, and let g E G . I f m E Mg n ChEG\{,} Mh, then m = CllEG\{g} mh with mh E Mh. Apply p t o both sides o f this equality, and then apply inn @ p , t o both sides t o obtain m @ 1 = 0 , where p, : K [ G ] --+K denotes the linear map defined by p Y ( C h E G k h h ) = k g . This implies that m = 0 , hence M = BgEG M,. Conversely, if M = BpEG M,, then clearly p is a well-defined linear map. I t can be easily checked that the map p endows M with a structure o f right K[G]-comodule, and any right K[G]-comodule is a G-graded K-vector space (see e.g., Montgomery [82, Example 1.6.71). 0 P~orosrrro 10.2.14. ~ The following assertions hold for a group G and a K-vector space A . ( 1 ) A is a right K[G]-comodule algebra if and only if A is a G-graded K -algebra. ( 2 ) A is a Galois K[G]-object if and only if A is a strongly G-graded K-algebra and A, = K . PROOF. For ( 1 ) use Lemma 10.2.13, and for ( 2 ) see Caenepeel [42, Proposition 8.2.11 or Montgomery [82, Theorem 8.1.71. 0
W e are now in a position t o present the connection between Cogalois field extensions and Hopf algebras.
COROLLARY 10.2.16. The following assertions are equivalent for a finite radical field extension E / F . ( 1 ) E I F is a Cogalois extension. ( 2 ) E is a Galois F [ C o g ( E / F ) ] - o b j e c via t the comodule structure given by the linear map E ---t E @F F [ C o g ( E / F ) 1, x I+ x 8 ?j, for all x E Fg and g E T ( E / F ) . PROOF. T h e result follows at once from Theorern 10.2.15, since an extension E / F is Cogalois i f and only if it is T ( E / F ) - K n e s e r ,and C o g ( E / F ) 0 =T(E/F)/F*.
Kext, we will present the concept o f H-Galois extension, as well as its connection with that o f Galois H-object.
DEFINITION 10.2.17. Let A be a right H-comodule algebra with structure nrap p : A -+ A @ H . W e say that ACO"c A is an H-extension. The H-extension ACO"c A is said to be an H-Galois extension if the map p :A H ---+ A @ K H , a @ b I+ abo @ bl 0 is bijective.
@ A r o ~
THEOREM 10.2.15. The following assertions are equivalent for a finite G-radical field extension E / F with G / F * finite. ( 1 ) E I F is a G-Kneser extension. ( 2 ) E is a Galois F [ G / F *]-object via the comodule structure given by the linear nrap E --+ E @ p F [ G / F * ] ,x I+ x 8 ?j, for all x E Fg and g E G . PROOF. ( 1 ) ==+ (2): By Proposition 10.1.6 (4), E / F is a strongly G/F*-graded extension via the decomposition E = E?, with Q = Fg for every g E G . Since = F 1 = F , we deduce by Proposition 10.2.14 ( 2 ) that E is a Galois F[G/F*]-object.
( 1 ) : B y Lemma 10.2.13 and Proposition 10.2.14 ( 2 ) , t h e exten(2) sion E / F is strongly G/F*-graded. Observe that U r l ( E )= G . Now, apply Proposition 10.1.6 ( 6 ) t o deduce that E / F is G-Kneser. 0
LEMMA 10.2.18. Let H be a Hopf algebra over K , and suppose that the antipode of H is bijective. Then, tlre following statements hold for a right H -comodule algebra A . ( 1 ) A is a Galois H-object. = K , and the linear map A 8 A -+ A @ H , a 8 b I+ C a b o @ b l ( 2 ) ACoZ1 is bijective. PROOF. See Caenepeel [42, Theorem 8.1.61.
T h e next result establishes the connection between Galois H-objects and H-Galois extensions.
PRo~osr~roN 10.2.19. Let H be a Hopf algebra over K , and suppose that tlre antipode of H is bijective. Then, the following statements hold for a right H-conrodule algebra A . ( 1 ) If A is a Galois H-object, then K = ACO"C A is an H-Galois extension. ( 2 ) If A ~ O " c A is an H -Galois extension and A ~ O " = K , then 4 is a Galois H-object. PROOF. Apply Lemma 10.2.18.
262
Chapter 10
253
We end this section by examining the connection between classical Galois field extensions and H-Galois extensions. Recall that if H is a K-Hopf algebra and M is a left H-module, then the set M ~ : = { ~ ~I E hm=~(h)m V ,~ E H } is a K-vector subspace of M which is called the subspace of H-invariants of M . If H is a finite dimensional K-Hopf algebra, then the K-vector space H* = HornK (H, K ) has a canonical structure of a K-Hopf algebra which is called the dual Hopf algebra of the K-Hopf algebra H .
LEMMA 10.2.20. Let H be any finite dimensional K-Hopf algebra, and let A be any K-algebra. Then A is a right H-comodule algebra if and only if A is a left H*-module algebra. Moreover, in this case we also have that A//* = A C O / ~
PROOF.See e.g., D&c$lescu, Niktikescu, and R.aianu [54, Proposition 6.2.41 or Montgomery [82, Lemma 1.7.2 and p. 411. i l In particular, if EIF is any field extension, and G G a l ( E / F ) is any finite group of F-automorphisms of E , then it is easily verified that E is a left F[G]-module algebra. Consequently, by Lemma 10.2.20, E is a right F[G]*-comodule algebra and
1. Let K I F be a field extension, let u E K be an algebraic element over F of degree n > 1, and set E = F ( u ) and Ek = F u k for every k E Z. Ek defines EIF (a) Prove that the decomposition E = CkEZ as a Z-Clifford extension if and only if urn f i ! F for every m E N*. (b) For every k, 0 k < n set Ez = F u k . Prove that the decomposition E = $ r-Z,,e defines EIF as a strongly Z , graded extension if and only if u" = a E F , or equivalently, Min(u, F ) = X" - a.
<
<
2. (Stefan [99]). Show that the following assertions hold for any rClifford extension EIF and any A , , A2 I?. (a) EAIEA, =E A ~ A ~ . (b) EA,n En, = EA, na, if EIF is a strongly I'-graded extension.
<
PROPOSITION 10.2.21. The following assertions are equivalent for a field extension EIF and a finite group G with G Gal(E/F) . (1) EIF is a Galois field extension with G = Gal(E/F). (2) E is a Galois F[G]*-object. (3) F c E is an F[G]*-Galois extension and F = Fix(G).
3. (Stefan [99]). Let E/F be a r-Clifford extension, let K be an intermediate field of E I F , and set I'K= { 7 r I Er K }. Prove that the following statements hold. (a) rK I ' . (b) Er, C K . (c) If A r a n d K = E A , then Er, = E A .
< <
<
PROOF. Denote K = Fix(G) and H = F[G]*. Then, as we observed =K. above, we have EC0" (2): If E/F is a Galois extension with G = Gal(E/F), then (1) we have F = K = E""". Now continue as in D&c$lescu, Kiistiisescu, and Raianu [54, Example 6.4.3 (I)] (see also h h t g o m e r y [82, 8.1.21). (2) =+ (1): If E is a Galois F[G]*-object, then by definition we have F = K = E ~ ~ "By . the Artin Theorem (see 1.2.9), E / K is a Galois field extension and G = Gal(E/K) = Gal(E/F), as desired. (2) w (3) follows from Proposition 10.2.19.
4. (Stefan [99]). Show that the following statements hold for a strongly I'-graded extension E I F , with I' a group of order 4 and Char(F) # 2. (a) If r is a cyclic group, 7 is a generator of r , x E E;, and z E E* is such that z' E F , then either z E F U Ern, or there 2 = -a2 and z = b(ax + x3) for exists an a E F* such that ' some b E F . (b) If r IZ2 x Z 2 and z E E* is such that z' E F , then z E U1&(E).
5. (Stefan [gel). Let E/F be a separable I'-Clifford extension, with I' a finite group of exponent n , and let p be a prime number dividing II'l. Assume that for every q E P , and for every cyclic subgroup A of order q of I' one has [EA: F] = g. Prove the following statements. ' . (a) [ EI., : F ] = d for any subgroup ro of order p2 of I
Chapter 10
10.4 Bibliographical corr~rr~er~ts to Chapter 10
255
(b) Whenever A l and A2 are subgroups of r such that lAl 1 = p, lA2l = p' for some r E W , A l & A2 and EAI= Ea,, then necessarily A = A2.
10.4. Bibliographical comments to Chapter 1 0
6. (Stefan [99]). Prove that the following assertions are equivalent for a separable I?-Clifford extension EIF with I? a finite group of exponent n. (a) EIF is a strongly r-graded extension. (b) For every p E P, and for every cyclic subgroup A of order p of r one has [EA: F ] = p . (c) For every odd prime p, C, E Uh(E) b E F , and 1+ E Uh(E) C4 E F . F. F and (1 p q ( E ) )n U"(E) (d) pP(E) n U"(E)
c4
Section 10.1. The concepts of a Clifford system and a Clifford extension were invented in 1970 by Dade [50], [51]. Dade also introduced ten years later the concept of strongly I?-graded ring [52]. A good account of graded rings and modules can be found in Kiistiisescu and Van Oystaeyen [84]. The interpretation of finite G-radical, G-Kneser, and G-Cogalois extensions via Clifford and strongly graded extensions is due to Stefan [99]. The results of this section are taken from Albu [14], where most of Stefan's results were generalized from finite t o infinite field extensions. Section 10.2. We followed Sweedler [102], Montgomery [82], Caenepeel [42], D&c%lescu, N&t&escu, and R.aianu [54] for the standard concepts ancl facts of Hopf algebras presented in this section. The connection between Cogalois extensions and Hopf algebras is mentioned in passing in Greither and Harrison 1631. The explicit connections provided by Theoremcl0.2.15 and Corollary 10.2.16 is due to Albu [14]. The concept of Galois H-coobject is defined and investigated in the theory of Hopf algebras and is the formal dual of the Galois H-object (see, e.g., Caenepeel [42, Section 8.71). It is not clear how this concept is related t o that of Cogalois field extension.
7. (Stefan [99]). Prove that the following assertions are equivalent for a separable r-Clifford extension E / F with r a finite group of
exponent n. (a) E/F is a r-Clifford-Cogalois extension. (b) pp(E) = p p ( F ) for all p E PrL. 8. (Stefan [99]). Let EIF be a finite strongly I"-graded extension, such that I' is the internal direct sum of two subgroups rl and I ' z . Prove the following assertions. (a) E and Er, QDF Er., are isomorphic F-algebras. (b) If E/F is additionally an Abelian extension, then Gal(E/F) Z Gal(E1,,I F ) x Gal(EI., I F ) . 9. (Stefan [go]). Let EIF be a I?-Clifford-Cogalois extension, with I? a finite group of exponent n. Assume that ; ,El and let A r be such that EA = F(CrL). Prove that E/F is a Galois extension and that there exists an exact sequence of groups:
cn
<
10. (Stefan [99]). Show that if EIF is any Galois r-Clifford-Cogalois extension, with I' a finite group of exponent a prime number, then G a l ( E / F ) 2 r . (Hint: Use Exercise 9.) 11. (Stefan [99]). Show that if E/F is any Galois r-Clifford-Cogalois extension, with l? a finite group of exponent n , and p,,(E) = /.L,,(F), then Gal(E/F) I?. (Hint: Use Exercise 9.)
Part 2
INFINITE COGALOIS THEORY
CHAPTER. 11
INFINITE KNESER EXTENSIONS

The aim of this chapter is to introduce and investigate one of the basic concepts of the Infinite Cogalois Theory, namely that of infinite Kneser extension. Infinite G-Kneser extensions are introduced and characterized in Section 11.1. One says that an extension E I F , which is not necessarily finite, is G-Kneser if G is a subgroup of E* containing F* such that E = F ( G ) and the factor group G I F * is a torsion group having a set of representatives which is linearly independent over F. We show that the Kneser Criterion, which characterizes finite separable G-Kneser extensions, can be generalized t o infinite extensions. Section 11.2 introduces and investigates infinite strongly G-Kneser extensions. The key result of this section is Theorem 11.2.4 which, roughly speaking, states that whenever two of the extensions K I F , E I K , EIF in a tower of fields F E K C E are Kneser, then so is the third one.
11.l. Infinite G-Kneser extensions The aim of this section is to introduce and characterize the notion of infinite G-Kneser extension. We also present the infinite variant of the Kneser Criterion. By Corollary 2.1.10, a finite G-radical extension EIF is G-Kneser if and only if there exists a set of representatives of G I F * which is linearly independent over F . This property suggests we generalize the concept of G-Kneser extension t o arbitrary field extensions as follows:
i
DEFINITION 11.1.1. An extension E I F , which is not necessarily finite, is said to be G-Kneser if it is a G-radical extension such that there exists
260
Chapter 1 1
1 1 . I Infinite G-Kneser extensions
26 1
a set of representatives of the factor group G I F * which is linearly independent over F . The extension EIF is called Kneser if it is G-Kneser for some group G. 0 The next result is a restatement of Proposition 2.1.8 in terms of infinite and finite Kneser extensions.
Q ( H ) / Q would be H-Kneser by Proposition 11.1.2, which contradicts ExCI ample 2.1.15 (1) since Q(H) = Q(C3). Next, we state and prove the infinite variant of the Kneser Criterion (Theorem 2.2.1).
PROPOSITION 11.1.2. The following assertions are equivalent for an arbitrary G-radical extension E/ F . (1) EIF is a G-Kneser extension. (2) Every set of representatives of G I F * is linearly independent over F . (3) Every set of representatives of G I F * is a vector space basis of E over F . (4) There exists a set of representatives of G I F * which is a vector space basis of E over F . (5) Every subset of G consisted of elements having distinct cosets in the group G / F * is linearly independent over F . (6) Every finite subset (xl , . . . ,x,) E G such that 6 # 6 for every i, j E (1,. . . , n ) , i # j , is linearly independent over F. (7) For every subgroup H of G containing F*,the extension F ( H ) / F is H-Kneser. (8) For every subgroup H of G such that F*< H and HIF* is a finite 0 group, the finite extension F ( H ) / F is H -Kneser. COROLLARY 11.1.3. If EIF is a G-Kneser extension, then I G I F * I = [ E : F]. PROOF. Since El F is a G-Kneser extension, there exists a set of representatives of G I F * which is linearly independent over F . Thus I G I F * 1 [ E : F]. By Lemma 2.1.6 we also have the opposite inequality I G I F * I 2 0 [ E : F],and consequently I G I F * I = [ E : F ] .
THEOREM 11.1.5 (THE INFINITE KNESERCRITERION). Let EIF be an arbitrary sepanzble G-radical extension EIF. Then, the following assertions are equivalent.
(1) E/F is a G-Kneser extension. 6 E F , and 1 f C E G (2) F& every odd prime p, 5, E G C4 E F . (3) p, (G) = /J,(F) for every odd prime p, and I f (4 E G C4 E F.
PROOF. (1) ===+ (2): Suppose that the extension EIF is G-Kneser. If C, E G, where p is an odd prime, then set H = F*(C,), and observe that Cp E H, F* 6 H 6 G, and H I F * is a finite group. By Proposition 11.1.2, the finite extension F ( H ) / F is H-Kneser. Now apply the (finite) Kneser Criterion (Theorem 2.2.1) t o deduce that C, E F . If 1 f C4 E G, then use a similar argument, by considering as H the subgroup F * ( l C4) of G.
<
The converse of Corollary 11.1.3 is not true, as the following example shows.
(1): Suppose that (2) is true. We will show that condition (8) (2) in Proposition 11.1.2 is verified, which will prove that the extension EIF is G-Kneser. Let H G with F* G and H I F * finite. Then F ( H ) / F is a finite extension. In order t o show that this extension is H-Kneser, we will use again the (finite) Kneser Criterion. If Cp E H , where p is an odd prime, then Cp E G, and by assumption it follows that Cp E F . If I f C4 E H, then 1 f C4 E G, and by (2) it follows that C4 E F . So, the finite extension F ( H ) / F is H-Kneser, and by Proposition 11.1.2 the extension EIF is GKneser. (3) as in the proof of Theorem 2.2.1. 0 (2)
<
<
EXAMPLE 11.1.4. Let p , , . . . ,p,,, . . . be the sequence of all positive prime numbers, and set E = Q(C3, @,. . . . . ). We have
,a,.
fi, . . . ,&, . . . )-Kneser. InHowever, the extension E l Q is not Q (6, deed, set G = Q ((3, . . , &, . . . ) and H = @ (C3). If the extension E / Q were G-Kneser, then it would follow that the finite extension
a,.
EXAMPLE 11.1.6. Any G-radical extension E I F , with E a subfield of the field R of real numbers is a G-Kneser extension since it verifies condition (2) of Theorem 11.1.5. In particular, if A C R is any set of real roots of positive rational numbers, then the extension Q(A)/Q is QY (A)Kneser. 0
Chapter 1 1
11.2 Infinite strongly Kneser extensions
263
11.2. Infinite strongly Kneser extensions
The aim of this section is to introduce the infinite variant of the concept of strongly Kneser extension discussed in Section 4.2, and t o investigate its main properties. The key result of this section is Theorem 11.2.4 which, roughly speaking, states that if in a tower of fields F K & E two extensiorls are Kneser, then so is the third one. Throughout this section, E/F will denote a fixed extension and G a group such that F* G E * . We shall use in the sequel the following notation:
Observe that any set of representatives of G / H will be also a set of representatives of (K*G)/K*, because of the group isomorphism
induced by the composition of the canonical maps Let { gi I i E I ) be a fixed set of representatives of G / H , and assume that {gi 1 i E I ) is linearly dependent over K . Then, there exists a finite subset {gil,...7gi,n) of { g i l i I ) and P I , . . . , p , E K * such that For simplicity, set gi, = gl ,. . . ,gin, = 9,. By hypothesis, gi H # gj H for every i # j in I. Since E = F ( H ) = F[H], for each i = 1, . . . ,m we can write
< <
G={HIF*<H<G),
E := Intermediate ( E I F ) = { K I F E K , K subfield of E ). LEMMA 11.2.1. Let E/F be a G-Kneser extension. Then, the extension F ( H ) / F is H-Kneser and F ( H ) n G = H for every H 6 G.
PROOF.The extension F ( H ) / F is H-Kneser by Proposition 11.1.2. The proof of the equality F ( H ) n G = H is literally the same as that in the case of finite extensions (see Proposition 2.1.11), and therefore is 0 omitted. The next result, which is fundamental in generalizing the Finite Cogalois Theory t o infinite extensions, is the infinite variant of Proposition 4.2.1. Its proof is different from and more complicated than the one in the finite case.
with ni E N*, Xij E F * , Itij E H , and hijF* # hikF* for all 1 ni. T ~ u s !
7I1 7IL
71;
<j
<
NOW, hijgiFt # hikgiF* for j # k, for otherwise it would follow that hijF* = hikF*, which is a contradiction. On the other hand, we have for any i # k, for otherwise, it would follow that gig;' E H , i.e., giH = g k H for i # k, which is a contradiction. Consequently, { hijgi I 1 i m, 1 j ni ) is a subset of G having distinct cosets modulo F*. Since E/F is G-Kneser, these elements are linearly independent over F by Proposition 11.1.2. Thus, Xij = O for every 1 i m, 1 j ni, which is a contradiction. So, {gi ( i E I ) is linearly independent over K , i.e., E / K is K*G-Kneser.
PROPOSITION 11.2.2. The following assertions are equivalent for a GKneser extension E/F and an intermediate field K of EIF.
(1) K / F is H-Kneser for some H E (2) K I F is K * n G-Kneser. (3) E / K is K*G-Kneser.
< <
< <
G.
< <
< <
PROOF. (1) (3): Suppose that K I F is H-Kneser for some H E G. Then I< = F ( H ) and F ( H ) n G = H by Lemma 11.2.1. In order t o show that E / K is K*G-Kneser it is sufficient, by definition, t o prove that there exists a set of representatives of (K*G)/K* which is linearly independent over K .
(3) Then
(2): Suppose that E / K is K*G-Kneser, and set K * n G = H .
(K*G)/K* G / ( K * n G) = G / H . We know that F ( H ) / F is H-Kneser by Proposition 11.1.2, so it will be sufficient t o show that I< = F ( H ) .
264
Chapter 1 1
11.2 Infinite strongly Kneser extensior~s
265
Consider the tower of fields and set L = F ( H ) . Since E I K is K*G-Kneser by hypothesis, any set of representatives of is a basis of E over K . But L I F is H-Kneser, hence using the implication (1) (3) we already proved, we deduce that E I L is L*G-Kneser; so, any set of representatives of G / H is a basis of E over L because Now, consider the tower of fields which has the property that there exists a basis of E over K which is also a basis of E over L. We claim that K = L, which will finish the proof. Indeed, assume that K # L, and take a basis { xi 1 i E I ) of E over K which is also a basis of E over L. Let { yj I j E J ) be a basis of K over L with yj, = 1 and yj, = y, where jl and j2 are two distinct fixed elements in J and y is a fixed element in K \ L. Then, by the Tower Law,
E over K since E I K is a K*G-Kneser extension. So, if { h i 1 i E I ) is a set of representatives of H I F * and { gj I j E J ) is a set of representatives of G I H , then, by the Tower Law, { higj 1 (i,j ) E I x J ) is a basis of E over F . On the other hand, we claim that { higj I (i, j ) E I x J) is a set of representatives of G I F * , which will finish the proof. Indeed, let g E G; * h i F * for then g H = g j H for some j E J , i.e., gg;l E H . Then g g y l ~ = some i E I , i.e., g F * = higjF*. Now,
This finishes the proof.
Propositions 11.2.2 and 11.2.3 can be reformulated together as follows:
THEOREM 11.2.4. Let F C K E E be a tower of fields, and let G be a group such that F* G E * . Consider the following statements: (1) K I F is K * n G-Kneser. (2) E I K is K*G-Kneser. (3) EIF is G-Kneser. 0 Then, any two of the assertions (1)-(3) imply the remaining one.
< <
B := { zipj 1 (i, j ) E I x J )
is a basis of E over L, in particular, the subset of B is linearly independent over L. But { xi ( i E I ) is a basis of E over L, hence, for a fixed io E I, xi, y is an Llinear combination of { xi 1 i E I ), which contradicts the linear independence of C over L. This proves our claim. (2) (1) is obvious.
Now, we are in a position to introduce a basic concept of the Infinite Cogalois Theory.
DEFINITION 11.2.5. An extension EIF is said to be strongly G-Kneser if it is a G-radical extension such that for any intermediate field K of E I F , the extension E I K is K*G-Kneser, or equivalently, by Proposition 11.2.2, the extension K I F is K * n G-Kneser. The extension EIF is called strongly Kneser if it is strongly G-Kneser for some group G. 0
Clearly, any strongly G-Kneser extension is G-Kneser, but not conversely, as Example 4.2.4 shows. h'ow consider an arbitrary G-radical extension E I F . Recall that we introduced in Section 4.1 the following notation:
PROPOSITION 11.2.3. Let E/F be an extension, and let G be a group such that F* G E * . If K is an intemnediate field of EIF such that K I F is K * n G-Kneser and E I K is K*G-Kneser, then EIF is G-Kneser.
< <
PROOF. First of all, observe that by the proof of Proposition 4.2.2, EIF is a G-radical extension.
Set H = K * n G. By Proposition 11.1.2, any set of representatives of H I F * is a basis of K over F since K I F is a K *nG-Kneser extension, and any set of representatives of ( K * G ) / K * Z G / ( K * n G ) = G I H is a basis of
& = { E lF C E , E s u b f i e l d o f E ) ,
266
Chapter 1 1
11.4 Bibliographical corr~rrients to Chapter 11
267
We have seen that the maps cp and Y E - G

11,
$ J
define a Cogalois connection
Let E,. = Q(c0s(a/2~+')),r E K , and Em = Ur2, E,. Prove that E,/Q is a Galois extension of infinite degree. Let
i
___t
which we have called the standard Cogalois connection associated with EIF. The next two results are the infinite variants of Theorem 4.2.7 and Proposition 4.2.8, respectively. Their proofs are literally the same as in the case of finite extensions, and therefore will be omitted.
t R. r 3 2. vr = 1
+,
7
Fr = Q(vr),
r radicals
and F , = Urz, Fr. Prove that F,/Q infinite degree:
is a non Galois extension of
THEOREM 11.2.6. The following assertions are equivalent for an arbitrary G-radical extension EIF. E/F is strongly G-Kneser. E/F is G-Kneser, and the map $ : G -+ E, $(H) = F ( H ) is surjective. E/F is G-Kneser, and every element of E is a closed element in the standard Cogalois connection associated with E I F . E/F is G-Kneser, and the map cp : E --+ G, p ( K ) = K n G is injective. E/F is a G-Kneser extension with G/F*-Cogalois correspondence. E/F is G-Kneser, and the maps - f G ' : E ---+ G, F ( - ) : G --+ I are isomorphisms of lattices, inverse to one another. 0
Prove that the following assertions hold for a field extension E / F . (a) Let G be a group with F* ,< G E * . If E/F is an arbitrary G-Kneser extension, then E/F is a strongly G/F*-graded extension. (b) Conversely, if E/F is a strongly r-graded extension for some torsion group I ' , then there exists a group G such that F* G E*, I ' 2 G / F * , E/F is G-Kneser, and U'YE) = G. (c) If EIF is a strongly G/F*-graded extension, where G is a group such that F* 6 G T ( E / F ) and U h ( E ) C G, then E/F is a G-Kneser extension.
<
<
<
<
P~orosrrro 11.2.7. ~ The following assertions hold for any strongly G-Kneser extension E/F and for any intermediate field K of E / F . (1) K I F is strongly K * n G-Kneser. 0 (2) E / K is strongly K*G-Kneser.
Prove that the following assertions are equivalent for a G-radical field extension E I F . (a) E/F is a G-Kneser extension. (b) E is a Galois F[G/F*]-object via the cornodule structure given by the linear map E + E @ F F[G/F*], x I+ x 8 $, for all x E F g and g E G.
11.4. Bibliographical comments to Chapter 11 Section 11.1. The concept of an infinite Kneser extension is due t o Albu and Tena [25]. The contents of this section follows closely the presentation in Albu and Tena [25]. Section 11,2. The contents of this section closely follows the presentation in Albu and Tena [25] where the notion of an infinite strongly Kneser extension was introduced and investigated.
1. Let E/F be an extension, which is simultaneously G-Kneser and H-Kneser. Prove that if H G, then H = G.
<
2. Let F 5 K that F* G extension and that E/F is a
c E be a tower of fields, and let G be a group such < < E * . Assume that K I F is a strongly K * n G-Kneser
E / K is a strongly K*G-Kneser extension. Is it true strongly G-Kneser extension?
CHAPTER 12
INFINITE G-COGALOIS EXTENSIONS

In this chapter we generalize the concept of finite G-Cogalois extension t o infinite extensions, and show that almost all of the results of Chapter 4 concerning finite G-Cogalois extensions also hold for infinite G-Cogalois extensions. In Section 12.1 we introduce the concept of an arbitrary G-Cogalois extension. The main result of this section is the General Purity Criterion (Theorem 12.1.4) which characterizes G-Cogalois extensions in terms of PG-purity. This result is applied t o show that a separable G-Kneser extension EIF is G-Cogalois if and only if the group G I F * has a prescribed structure. We will use this criterion in the next chapter t o provide large classes of infinite G-Cogalois extensions. Also, we particularize these results to bounded G-Cogalois extensions. Section 12.2 deals with infinite Cogalois extensions. We show that rnost of the results on finite Cogalois extensions established in Chapter 3 also hold for infinite Cogalois extensions.
12.1. The General Purity Criterion and its applications
The most interesting G-radical extensions EIF with G/F*-Cogalois correspondence are those which are also separable. The finite extensions of this type were called G-Cogalois extensions in Section 4.3. We will preserve this terminology also in the case of infinite extensions. We characterize these extensions in terms of PG-purity, and show that, as in the case of finite extensions, they can be described within the class of separable GKneser extensions by the structure of their uniquely determined Kneser groups.
DEFINITION 12.1.1. An extension, which is not necessarily finite, is 0 said to be G-Cogalois if it is separable and strongly G-Kneser.
269
270
Chapter 12
12.1 The General Purity Criterio~iand its applicatior~s
27 1
The next result shows that the property of an extension being GCogalois behaves nicely with respect to subextensions and quotient extensions
PROPOSITION 12.1.2. Let EIF be a G-Cogalois extension. Then, for every intermediate field K of E I F , the following assertions hold. (1) K I F is K * n G-Cogalois. (2) E I K is K*G-Cogalois.
PROOF.The result follows at once from Proposition 11.2.7.
Now suppose that 4 E PG and C4 E E\F. Then necessarily Char(F) # 2, and there exists 94 E G with o r d ( 5 ) = 4 in the quotient group G I F * . As (7) in Theorem 4.3.2 one deduces that in the proof of implication (2) C4 E G. Denote K = F ( ( l C4)g4). Since the extension EIF is strongly ~ K*G, 4 ~ ~ ~ G-Kneser, E / K is K*G-Kneser. But 1 C4 = (1 ~ . l ) E hence (4 E K by the Infinite Kneser Criterion (Theorem 11.1.5). Now continue as in the proof of implication (2) (7) in Theorem 4.3.2 t o deduce that C4 E F * . But this contradicts our initial assumption that <4 @ F . Consequently, E/F is PG-pure.
Recall that by P we have denoted the set of all strictly positive prime numbers, by P the set ( P\(2)) u {4), by ID, the set of all positive divisors of a given number n E N,and by P,, the set P n ID,. DEFINITION 12.1.3. Let Q be a nonempty subset of P . An extension
EIF is said to be Q-pure if p p ( E ) & F .
Note that the concepts of pure and n-pure extension are particular cases of this more general notion: an extension is pure precisely when it is P-pure, and n-pure when it is ?,-pure. Recall that for an arbitrary torsion group T we have denoted in Section the set { ord(x) I x E T ). 1.4 by 07, For any G-radical extension EIF we shall use throughout this chapter the following notation:
P,
:= P
n o,,,.. .
Note that when the G-radical extension EIF is n-bounded (see Definition 5.1.1), then OG = ID,, , hence PG = P n ID, = P,,. Next, we provide a purity criterion for arbitrary G-Cogalois extensions, which generalixes to infinite extensions the n-Purity Criterion (see Theorem 4.3.2).
(1): Assume that E/F is P,-:-pure, and let K be an arbitrary (2) subfield of E containing F . We are going t o show that K/F is K * n GKneser. To do that, we will use the Infinite Kneser Criterion cited above. Let p be an odd prime such that Cp E K * nG. Since C,P = 1, we deduce that either o r d ( c ) = 1 or ord(&) = p. If o r d ( c ) = 1 we have Cp E F , as desired. If ord(&) = p, it follows that p E PG since Cp E G, hence C, E F by PG-purity. Now suppose that 1 C4 E K*n G , and set g = 1 C4. Then necessarily Char(F) # 2, for otherwise, it would follow that C4 = 1, hence 1 C4 = 0 E K * n G, which is a contradiction. " -4 E F* one has m E {1,2,4). If m = 1 Set m = ord(g"). Since g then 1 Cq E F * , hence C4 E F . If m = 2 then g " 2 2 C q E F * , hence C4 E F since Char(F) # 2. If m = 4 then 4 E P G . But 1+C4 E K * n G & E implies that C4 E E , hence C4 E F by Pc-purity. Thus E/F is strongly G-Kneser, and since it is also separable, we conclude that EIF is G-Cogalois. The proof is complete. 0
THEOREM 12.1.4 ( T H E GENERAL PURITY CRITERION). Let EIF be an arbitrary G-radical separable extension. Then, the following assertions are equivalent. (1) E/F is G-Cogalois. (2) E/F is PG -pure.
PROOF. (1) ===+ (2): Suppose that the extension EIF is G-Cogalois and let p be an odd prime in P G . By definition, there exists gp E G with ord($,) = p in the quotient group G / F * . Continue as in the proof of implication (2) a (7) in Theorem 4.3.2 to deduce that Cp E F .
REMARK 12.1.5. The proof of the implication (2) =j (1) in Theorem 12.1.4 shows that any separable G-radical extension E/F with PG= 0 is G-Cogalois. Notice that PG= 0 if and only if either G / F * is a group of exponent 2 or G = F * . 0 Theorem 12.1.4, one of the key results in the Infinite Cogalois Theory, has many consequences. First, we will show that the property of an extension being G-Cogalois behaves nicely with respect to directed unions. This explains why most results from Finite Cogalois Theory also hold for infinite extensions.
PROPOSITION 12.1.6. Let EIF be an extension. Assume that E = U a E l E, is a directed union of a family of internaediate fields E, of E/F such that every extension E,/F is G,-radical and G = UuEIG, is a
272
Chapter 12
12.1 The General Purity Criteriori arid its applications
273
directed union of subgroups G,. Then E/F is G-Cogalois if and only if E, IF is G, -Cogalois for every a E I. PROOF. Assume that E/F is G-Cogalois. Then E,/F is G n E,Cogalois for every a E I by Proposition 12.1.2 (1). Observe that G n E, = G n F(G,) = G, by Lemma 11.2.1. Conversely, assume that every E,/F is G,-Cogalois. Note that E/F is a separable G-radical extension. Let p E P G . Then, there exists gp E G such that ord(%) = p. Since G = UaE, G, we deduce that gp E Go for some B E I. Assume that E E . Then C, E E, for some y E I. Since G = Upel Gp is a directed union, there exists a E I such that E, C E, and Gs 5 G,. Then E E, and p E P G U . But E,/F is G,-Cogalois, E F by the General Purity Criterion (Theorem 12.1.4). Thus, hence EIF is PG-pure, hence it is G-Cogalois again by the same criterion.
Also, recall that for any extension EIF we have denoted by Cog(E/F) the torsion subgroup t ( E * / F * ) of the quotient group E * / F * , and by Cog2(E/F) the subgroup of Cog(E/F) consisting of all its elements of order 6 2.
THEOREM 12.1.8. The following statements hold for a separable GKneser extension E/ F . (1) Assume that 4 E P G . Then E/F is G-Cogalois if and only if
cp
cp
cp
G I F * = ( $ tp(Cog(EIF)))
pEPc
$ Cog, ( E I F ) .
I F '\ P G , so
Another application of the General Purity Criterion is the determination of the structure of the group G / F * for an arbitrary G-Cogalois extension which generalizes the results of Section 4.4 established for finite G-Cogalois extensions. To do that we need the next result which is the infinite variant of Lemma 6.1.6. LEMMA 12.1.7. Let E/F be a G-Cogalois extension, and let x E E* be such that xm E F for some m E W . Suppose that one of the following two conditions is satisfied. (1) P r n C PG. (2) p,,(E) E F (in particular, this holds if C,, E F ) . x E G. Then, we have F ( x ) E E PROOF. Suppose that F ( x ) C E. Set K = F ( x ) and H = F*(x). If k = ord(Z), then clearly k I nt and k = exp ( H I F * ) . Let p E P k . Then p l m , hence p E P,, 5 PG if condition (1) is satisfied. We deduce that p p ( K ) C p p ( E ) C F since EIF is PG-pure by Theorem 12.1.4. If condition (2) is satisfied, then we have p p ( K ) C pp(E) 5 p,(E) 5 F . Observe that K is a finite H-radical extension, hence, in both cases, K / F is a H-Cogalois extension by Theorem 4.3.2. But, the finite extension K / F is also G n K*Cogalois by Proposition 4.3.5 ( I ) , so H = G n K * by Corollary 4.4.2. Thus, x E G, as desired. Note that the result holds even if P,, = 0.The other implication is obvious. 0 R.ecal1 that if A is a multiplicative group with identity element e, then for every p E P we have denoted by tp(A) the pprimary component of A.
PROOF. (1) "=J": Clearly tp(G/F*) = 0 for all p E
Using Lemma 12.1.7 we will show that for all p E (PG \ (4)) U (21, which will imply the desired equality
So, let p E (PG \ (4)) U (21, and let xp E E* with E tp(Cog(E/F)). Then x$ E F for some s E N. We can assume that s ) 2. If p is odd then Pp.q = P n Q,.. = {p) P G , and if p = 2 then p2' = P n m2., = (4) p G . Now apply Lemma 12.1.7 to conclude that xp E G. Consequently
c c
The other inclusion is obvious.

*?l
hence
cP
: Let p
E PG\ (4) and assume that Cp E E . Then = 1, E G / F * . Thus, 5, E G. But E/F E tp(Cog(E/F)), and so
c;
274
Chapter 1 2
12.1 The General Purity Criterion and its applicatior~s
278
is G-Kneser by hypothesis, so by the Infinite Kneser Criterion (Theorem E F. 11.1.5) we deduce that E E and Char(F) # 2, then 1 E By assumption, 4 E PG.If t2(Cog(E/F)) & G / F * since (1 + c4)? = -4, and so, 1 E G. Again E F. If Char(F) = 2, by the Infinite Kneser Criterion, we deduce that then = 1 E F . This shows that E/F is PG-pure. Now apply Theorem 12.1.4 t o conclude that E/F is G-Cogalois.
cP
c4
cq
c4
+ c4
+ c4
finite extension K / F is simultaneously GI-Cogalois and HI-Cogalois, and according to Corollary 4.4.2 it follows that G' = HI. Since x E GI, we deduce that x E HI, so x E H . Hence G & H. The proof of the inverse inclusion H E G is similar. 0 In view of Theorem 12.1.10, the group G of any G-Cogalois extension, finite or not, is uniquely determined. So, this leads t o the following definition. DEFINITION 12.1.11. If E/F is a G-Cogalois extension, then the group G I F * is called the Kneser group of the extension E/F and is denoted by Kne(E/F). 0 Next, we particularize the General Purity Criterion (Theorem 12.1.4) we established for arbitrary G-Cogalois extensions to bounded G-Cogalois extensions. Recall that a G-radical extension E / F , which is not necessarily finite, is said t o be bounded if G / F * is a group of bounded order; in this case, if exp(G/F*) = n, one says that EIF is an n-bounded extension.
(2) "=J": As in case (1) one shows that the equality t,(GIF*) = t,(Cog(EIF)) holds for any (odd) prime p E PC. The hypothesis 4 # PC irnplies that G I F * contains no element of order 4, hence t2( G I F * ) is exactly the set of all elements of G I F* of order 2. E Cog2(G/F*). Then x i E F , hence Let 2.2 E E* with
<
p2=Pr1&=0cpG. Apply Lemma 12.1.7 t o conclude that x E G. Therefore and consequently
THEOREM 12.1.12 (THE INFINITE n-PURITY CRITERION). The following assertions are equivalent for a separable n-bounded G-radical extension E/F. (1) E/F is G-Cogalois. (2) E/F is n-pure.
"+=":
Proceed as in case (1).
REMARK 12.1.9. The condition that E/F is a G-Kneser extension in Theorem 12.1.8 cannot be dropped since, otherwise, E/F may not be G0 Cogalois as the example in R,emarks 4.4.4 (1) shows. The next result generalizes Corollary 4.4.2 t o extensions which are not necessarily finite.
PROOF. Since exp(G/F*) = n , we have OGIF* = ID7, by Remark 1.4.9, hence PC= P n O,,. = P n ID7, = P,,. Now apply the General Purity Criterion. 0 EXAMPLE 12.1.13. Let { p I , . .. ,pk,. . . ) be an infinite set of positive prime numbers, and set E = Q( . . ., . . . ). We claim that the . . . ,@, . . . ). infinite extension E / Q is G-Cogalois, where G = Q ( Indeed, observe first that E / Q is clearly a G-radical extension, and the group G / Q has exponent n. Next, the extension E / Q is pure since . .. , . . . ) & R, and so, a fortiori, it is n-pure. Q( Thus, E / Q is an infinite G-Cogalois extension by Theorem 12.1.12. In particular, the lattice of all its subextensions, that is, the lattice of all subfields of E , can be easily understood if we know how the subgroups of its Kneser group G / Q look. If n = 2, the Abelian group G / Q has a canonical structure of vector space over the field IF2 with 2 elements, and any of its subgroups is a vector subspace of G / p . Consequently,
m, m,
m,
THEOREM 12.1.10. Let E/F be an extension which is simultaneously G-Cogalois and H -Cogalois. Then G = H . PROOF. Let x E G be arbitrary, and set K = F ( x ) , GI = F*(x). Since the extension E/F is H-Cogalois it follows by Theorem 11.2.6 that K = F ( H 1 ) for some H ' with F* H' H . By Proposition 12.1.2 (I), the finite extension K / F is simultaneously K * nG-Cogalois and also K*n H-Cogalois. But K * n G = F(G1)n G = GI and K * n H = F ( H 1 ) n H = HI by Lemma 11.2.1. Consequently, the
m, s,
<
<
276
Chapter 12
12.2 Irlfir~ite Cogalois extensions
277
any subgroup of G / Q has a vector space basis over IF2. This can help t o describe effectively all the subfields of E . See also Exercise 7 for a more general case. 0
12.2. Infinite Cogalois extensions As we have noticed in Examples 2.1.15 (2), a finite extension E/F is Cogalois precisely when it is T(E/F)-Kneser. Since we already have the concept of G-Kneser extension defined for extensions which are not necessarily finite, it is natural t o use "Cogdois" to refer t o an arbitrary extension EIF if it is T(E/F)-Kneser. I11 this section we generalize, t o infinite extensions, most of the results we established in Sections 3.1 and 3.2 for finite Cogalois extensions. DEFINITION 12.2.1. A n extension E I F , which is not necessarily finite is said to be Cogalois if it is T(E/F)-Kneser. The next result is the infinite variant of the Greither-Harrison Criterion (Theorem 3.1.7).
(1): Assume that the extension E/F is radical, separable and (2) pure. We are going t o prove that the extension E/F is Cogalois, i.e., T ( E lF)-Kneser. Since E/F is a radical extension, it follows that E/F is also a T ( E / F ) radical extension. For every odd prime p we have p p ( T ( E / F ) ) pP(E), and, by purity, pp(E) = pp(F), SO p p ( T ( E / F ) ) = pp(F). If 1 f C4 E T ( E / F ) , then I fC4 E E , SO (4 E E , i.e., C4 E p4(E). By purity, p q ( E ) C pq(F), hence (4 E F. Since EIF is a separable extension, by the Infinite Kneser Criterion (Theorem 11.1.5) we deduce that the extension E/F is T(E/F)-Kneser, i.e., it is a Cogalois extension. 0 The next result is the infinite variant of Theorem 7.5.1.
THEOREM 12.2.3. Any Cogalois extension E/F is T(E/F)-Cogalois.

PROOF. In view of the Infinite Greither-Harrison Criterion, the radical extension E/F is separable and pure, so, a fortiori, Py,(fi;,,./l;)-pore.Thus, E/F is T(E/F)-Cogalois by Theorem 12.1.4. D An easy consequence of Theorems 12.2.2 and 12.2.3 is the following generalization of Proposition 3.2.2 and Theorem 3.2.3 from finite Cogalois extensions t o arbitrary Cogalois extensions.
THEOREM 12.2.2 ( T H EINFINITE GREITHER-HARRISON CRITERION). The following assertions are equivalent for an arbitrary extension E I F . (1) E/F is Cogalois. (2) E/ F is radical, separable, and pure.
PROOF. (1) ==+ (2): Suppose that E/ F is a Cogalois extension, i.e., a T ( E lF)-Kneser extension. In particular, E/ F is a T ( E lF)-radical extension, hence a radical extension. Now we show that the extension E/F is separable. It is sufficient t o prove that for any x E T ( E / F ) , the extension F ( x ) /F is separable. So, let x E T ( E / F ) be fixed, and set H = F * ( x ) and K = F ( x ) = F ( H ) . Since the extension E/F is T(E/F)-Kneser and H 6 T ( E / F ) , we deduce that the finite extension F ( H ) / F is H-Kneser by Proposition 11.1.2. Then, by Proposition 11.2.2, it follows that the extension K I F is K * n T ( E / F ) = T(K/F)-Kneser, i.e., K I F is a Cogalois extension. Hence K / F is a finite Cogalois extension, so it is separable, according t o the Greither-Harrison Criterion (Theorem 3.1.7) for finite extensions. Consequently, any element of K , in particular x, is separable over F , as desired. In order t o show that the extension E/F is pure, proceed exactly as in the proof of Theorem 3.1.7.
THEOREM 12.2.4. The following statements hold for an arbitrary Cogalois extension E/F. (1) The maps - n T ( E / F ) : & -+ C and F(-) : C -+ E are isomorphisms of lattices, inverse to one another, where & = { K I F C K , K subfield of E ) and C = { H I F * 6 H 6 T ( E / F ) ) . (2) For evey intermediate field K E & one has K = F ( T ( K / F ) ) . (3) For e v e y subgroup H E C one has C o g ( F ( H ) / F ) = H I F * . (4) For every intermediate field K of the extension E / F , E / K and K / F are both Cogalois extensions, and there exists a canonical group isomorphisnb
PROOF. (1) Since Cog(E/F) = T ( E / F ) / F * , C is canonically isomorphic t o the lattice Subgroups(Cog(E/F)). The result follows immediately frorn Theorem 12.2.3 and Theorem 11.2.6. (2) and (3) are consequences of (1) and of the following simple fact: for any H E C one has H = F ( H ) n T ( E / F ) = T ( F ( H ) / F ) .
278
Chapter 12
12.3 1t;xercises to Chapter 12
279
(4) By Theorem 12.2.3, the extension El F is T (E/F)-Cogalois hence it is strongly T(E/F)-Kneser. It follows that K I F is T ( E / F ) n K-Kneser. But T ( E I F ) n K = T ( K / F ) , hence K / F is T (K/F)-Kneser. This means precisely that K I F is a Cogalois extension. By Theorem 12.2.2, E/F is radical, separable and pure. Then clearly E I K is also radical, separable and pure. Again by Theorem 12.2.2, we conclude that E I K is a Cogalois extension. It remains only t o prove the isomorphism
Let Em = Ur2, E, be the directed union of subfields E, of W. Then E,/F is a Galois extension of infinite degree by Exercise 3, Chapter 11. We claim that the extension E,/F is not radical. Indeed, if it would be radical, then, since it is pure, it would be Cogalois by Theorem 12.2.2. Consequently, by Theorem 12.2.4 (4), any of its subextensions, in particular E 2 / F , is Cogalois, which contradicts Proposition 5.3.1 (e). , is a subfield of the maximal 2-primary Observe that the field E Abelian extension of Q contained in G
Since E / K is Cogalois, it is T(E/K)-Cogalois by Theorem 12.2.3. On the other hand, since El F is T ( E lF)-Cogalois, it follows by Proposition 12.1.2 (2) that E / K is T(E/F)Ke-Cogdois. Using Theorem 12.1.10 we obtain the equality T ( E / K ) = T ( E / F ) K * . Hence
Since the extension Q(<27.)/Q is (@ (C2..)-Kneserfor every r 2 1, it follows that the infinite Abelian extension / Q is ({ (2.. / r 2 1))-Kneser. However, its subextension E,/Q is not Kneser. 0
@lab
and we are done.
1. Prove that every separable 2-bounded G-radical extension is GCogalois. 2. Prove that a G-radical extension EIF is G-Cogalois if and only if every finite subextension K / F of E/F is G n K-Cogalois. 3. Let EIF be a G-Cogalois extension, and let x E T ( E / F ) with m = ord(2). Show that x E G if and only if P,, E PG. 4. Prove that the following statements hold for w sepamble n-bounded G-Kneser extension E/ F . ( I ) -4ssume that n f 2 (mod 4). Then E/F is G-Cogalois if and only if GIF* = (2) Assume that n only if GIF* =
PROPOSITION 12.2.5. Let EIF be a radical extension, and let K be an intermediate field of E I F . If the extensions E / K and K/F are both Cogalois, then EIF is also a Cogalois extension. PROOF. Since E I K and K I F are separable extensions, so is E I F . Since E / K and X/F are pure, so is also EIF by Proposition 3.1.5. But E/F is a radical extension by hypothesis, hence EIF is Cogalois by Theo17 rem 12.2.2.
EXAMPI,ES 12.2.6. (1) Proposition 3.2.6 shows that the result in Proposition 12.2.5 may fail if the extension E/F is not supposed t o be radical.
(2) Consider the extension Er/F discussed in Roemark 5.3.2, where Er = Q(O, ), F = Q, and
@ t,(Cog(E/F)).
PEP,,
2 (mod 4). Then E I F is G-Cogdois if and
t,(Cog(E/F)))
@ CoI?,(E/F).
pEP,.\{'L}
T
r radicals
280
Chapter 12
12..1 13ibliographical con~rr~er~ts to Chapter 12
28 1
5. Let EIF be a G-radical pure separable extension. Prove that EIF is simultaneously Cogalois and G-Cogalois, and Cog(E/F) = G I F * . (Hint: Adapt the proof of Corollary 4.4.3.)
11. Let E,. = Q(cos(n/2"+')), r E K , and Em = U, ), Cog (Em /Q) = { 7, 12. Let r E
E,. Show that
}.
/ l +
r
6. Let {pl,. . . ,pk,. . . } be an infinite set of positive prime numbers, .., . . ). Prove that let n E N, n 2 2, and set E = Q(
m,. a,.
PI, r
) 2, u,. = \
\
7
F,. = Q(u,.),
(Hint: Use Exercise 5.)
7. Describe effectively all the subfields of the field E considered in Exercise 6.

8. ( The Infinite Gay- Vdez Criterion). Prove that the following statements are ecluivalent for an extension EIF which is not necessarily finite. (1) EIF is a Cogalois extension. (2) E/F is radical, separable, and C2p # E \ F for every p E P. (Hint: Adapt the proof of Theorem 3.1.9, by using the Infinite Greither-Harrison Criterion and Lemma 3.1.2.) 9. Let EIF be a separable n-bounded G-radical extension satisfying the following condition: (*) pp(E) E F for any odd p E P , , and p4(E) E F whenever n is even. Prove that EIF is G-Cogalois, and G I F * = tp(Cog(E/F)), (Hint: Observe that the condition p 4 ( E ) C F for n even implies that Cog, (GIF*) = tz (Cog(E/F)). Then use Exercise 4.) 10. Let El 1 F and E2I F be Cogalois extensions such that their compositum ElE2IF is pure. Prove that ElE2IF is a Cogalois extension and that the following assertions are ecluivalent. (a) T ( E , / F ) n T(E,/F) = F * . (b) The canonical map
radicals and F , = Ur2, F,. Prove that the non Galois extension of infinite degree F m / Q (see Exercise 4, Chapter 11) is neither radical, nor Kneser, nor Cogalois. 13. With notation of Exercise 12, calculate Cog (F,/Q). 14. The concept of r-Clifford-Cogalois can be obviously defined a s in Definition 10.1.8 also for field extensions which are riot necessarily finite. Prove that the following statements hold for a field extension EIF. (1) If E/F is G-Cogalois for some group G with F* G E * , then EIF is a G/F*-Clifford-Cogalois extension. (2) Conversely, if EIF is a r-Clifford-Cogalois extension for some group r , then there exists a uniquely determined group G such that F* G ,< E*, I ' E G I F * , EIF is G-Cogalois, and U" (E) = G.
< <
<
15. Let EIF be a field extension which is simultaneously r-Clifford' and h Cogalois and A -Clifford-Cogalois. Prove that the groups I are isomorphic. 16. Prove that the following assertions are equivalent for a radical field extension E/F. (a) EIF is a Cogdois extension. (b) E is a Galois F[Cog(E/F)]-object via the comodule structure given by the F-linear map E --+ E @ F . F [ C o g ( E / F ) ] , x I+ x @g^, for all x E F g and g E T ( E / F ) .
defined as a ( x l F * , x2F*) = x1x2F*, 21 E E l , 2 2 E E2, is a monornorphism of groups. (c) The canonical map a in (b) is an isomorphism of groups. (d) The fields El and E2 are linearly disjoint over F . (e) El n E2 = F.
12.4. Bibliographical comments t o Chapter 12
Sect ion 12.1. The concept of infinite G-Cogalois extension was introduced Albu and Tena [25]. The General Purity Criterion (Theorem 12.1.4) and Theorern 12.1.8 giving the structure of the group G I F * of a G-Cogalois extension are due to Albu [B]. The uniqueness of the Kneser group of a
282
Chapter 12
G-Cogalois extension was established by Albu and Tena [25].
Section 12.2. The notion of infinite Cogalois extension, as well as the infinite variant of the Greither-Harrison are due to Albu and Tena [25]. The properties and examples of infinite Cogalois extensions presented in this section are taken frorn Albu [8].
CHAPTER 13
INFINITE KUMMER THEORY

The prototype of an infinite G-Cogalois extension is, by Kummer Theory, any infinite classical Kummer extension. In this section we show that the essential part of the Infinite Kurn~ner Theory can be immediately deduced from our Cogalois Theory using the Infinite n-Purity Criterion given in Section 12.1. Moreover, this criterion allows us t o provide large classes of infinite G-Cogalois extensions which generalize or are closely related t o classical infinite Kummer extensions: infinite generalized Kummer extensions, infinite Kummer extensions with few roots of unity, and infinite quasi-Kummer extensions. The prototype of an infinite Kummer extensions with few roots of unity is any subextension of R/Q of the form Q({ 1 i E I))/Q, where { a; 1 i E I) is an arbitrary nonempty set of strictly positive rational numbers. Notice that, in general, these extensions are not Galois if n 3 3. Placing classical infinite Kummer extensions in the framework of infinite G-Cogalois extensions allows us to derive, frorn the Infinite Cogalois Theory, results on infinite generalized Kummer extensions. In particular, we obtain results on infinite Kurnrner extensions with few roots of unity, which are very similar to the known ones for infinite classical Kummer extensions.
13.1. Infinite classical Kummer extensions
In this section we present as an imrnediate application of the Infinite n-purity Criterion the basic properties of infinite classical Kurnrner extensions.
sa Recall that R is a fixed algebraically closed field containing F a subfield; any considered overfield of F is supposed t o be a subfield of R.
283
284
Chapter 13
13.2 Irifiriite generalized Kurnrrier exterisioris
285
As in Section 7.1, for any nonempty subset A of F* and any n E N* we will denote by the subset of T ( R / F ) defined by
in R for which its Galois group is a group of exponent a divisor of n. It will be denoted in the sequel by F,O$ , or more simply, by Ftb. Taking A = F*in Theorem 13.1.2 (2) we obtain the next result, which is the Kummer Theory for the iniuimal classical n-Kummer extension Fzb of F. COROLLARY 13.1.3. Let F be any field and let n E W be such that and gcd(n, e ( F ) ) = 1 and p , ( R ) & F . Then, the maps H I+ F( *) E c, En n F* establish isomorphisms of lattices, inverse to one another, between the lattice of all subgroups H of F* containing F*" and the lattice 0 of all classical n-Kummer extensions E of F (contained in R). The remaining part of the Infinite Kumrner Theory, namely that concerning the description of the Kneser group of an infinite classical Kummer extension as the group of continuous characters of its Galois group, will be examined in Section 15.3.
LEMMA 13.1.1. Let EIF be a bounded separable G-radical extension, and let n E W be such that GR F * . If the extension E/F is n-pure, then EIF is G-Cogalois.
PROOF. Proceed exactly as in the proof of Lemma 7.1.4, but apply Theorem 12.1.12 instead of Theorem 4.3.2. 0 The next result essentially shows that the main part of Kummer Theory can be very easily derived from the Infinite Cogalois Theory.
THEOREM 13.1.2. Let E/F be a classical n-Kummer extension, with E = F(i/;i), where n E N* and 0 # A C F * . Then, the following assertions ,hold. (1) E/F is an F* ( n ) - ~ o ~ a l oextension. is (2) The maps H H F ( ;/77) and K c, KT' n ( F l n ( A ) ) establish isomorphisms of lattices, inverse to one another, between the lattice of all subgroups H of F*n(A) containing F*n and the lattice of all intermediate fields K of E / F . Moreover, any subextension K / F of E/F is a classical n-Kummer extension. (3) If H is any subgroup of F*"(A) containing F * n , then any set of representatives of the group is a vector space basis of F ( m ) over F , and [ F ( m ) : F ] = IH/F*rLI.
13.2. Infinite generalized Kummer extensions In this section we introduce and investigate the infinite variant of the concept of finite generalized Kummer extension discussed in Section 7.2. Since such extensions are bounded G-Cogalois extensions, we can easily deduce that the main properties they enjoy are very similar t o those of infinite classical Kummer extensions. DEFINITION 13.2.1. We say that an extension E/F is a generalized nKumrner extension, where n E W , if E = F ( B ) for some 0 # B E*, 0 with gcd(n, e ( F ) ) = 1, B" F , and prL(E)E F. Clearly, a generalized n-Kummer extension EIF is finite if and only if the set B in Definition 13.2.1 can be chosen to be finite. 0bserve that any classical n-Kummer extension F( F of exponent n is a generalimed n-Kummer extension, since if we denote B = then BrL= A & F * . Conversely, if A" = B , then .4 2 ;/B, but the inclusion may be strict.
m/~*
PROOF. Proceed as in the proof of Theorem 7.1.6 using the nbounded infinite variants of the involved results (e.g., Lemma 13.1.1 instead 0 of Lemma 7.1.4, Theorem 12.1.12 instead of Theorem 4.3.2, etc.). Let F be any field, and let n E W be such that gcd(n,e(F)) = 1 and prL(R) F ; these conditions mean precisely that CrL E F and ord(Cr,) = n in R*. In view of Theorem 7.1.2, any classical n-Kummer extension E of F , with E a subfield of R, has the form F ( i / ; i ) for some 0 # A E F * . Since obviously F( )/F is a classical n-Kummer extension, it follows that the set Knln of all classical n-Kumrner extensions of F contained in R, ordered by inclusion, has a greatest element, namely F ( w ) . The field F ( p ) is called the maximal classical n-Kummer extension of F contained in R; clearly, it is the greatest Abelian extension of F contained
n)/
a,
THEOREM 13.2.2. Let E/F be a generalized n-Kummer extension, with # B E E * , E = F ( B ) , Bn F , gcd(n,e(F)) = 1, and p n ( E ) E F . If we denote G = F * ( B ) , then the following statements hold. (1) The extension EIF is G-Cogalois.
286
Chapter 13
(2) The maps H I+ F ( n G ) and K I+ K" fl ( F a n ( B " ) ) establish isomorphisms of lattices, inverse to one another, between the lattice of all subgroups H of F * " ( B T 1 )containing F*" and the lattice of all intermediate fields K of E I F . Moreover, any subextension K I F of E I F is a generalized n-Kummer extension. ( 3 ) If H is any subgroup of F*n(B n ) containing F*n, then any set of representatives of the group ( n G ) / F * is a vector space basis over F , and : F ] = IH/F*"I. of
13.4. Infinite quasi-Kummer extensions
In this section we present the infinite variant of the concept o f finite quasi-Kummer extension discussed in Section 7.4.
~(mn~) [~(mn~)
DEFINITION 13.4.1. An extension E I F is said to be a quasi-Kummer extension if E = F ( B ) for some 0 # B c E * , and there exists n E K with B n F , gcd(n, e ( F ) )= 1, such that C, E F for e v e y p E p7,. THEOREM 13.4.2. Any quasi-Kummer extension F ( B ) / F with B as in Definition 13.4.1 is an F* ( B ) - Cogalois extension. PROOF. Let G = F * ( B ) . Then clearly G" C F * , and E = F ( G ) . So, E I F is a bounded G-radical extension. I f m = exp(G/F*),then m In. Since gcd(n, e ( F ) ) = 1 by hypothesis, it follows that gcd(m, e ( F ) ) = 1, and consequently, E I F is a separable extension by Corollary 5.1.3. For every p E P,, we have p p ( E ) pp (a) F , SO E I F is n-pure. By Lemma cl 13.1.1, it follows that F ( B ) / F is an F*(B)-Cogalois extension.
PROOF. Proceed as in the proof of Theorem 7.2.3.
13.3. Infinite Kummer extensions with few roots of unity
The infinite Kummer extensions with few roots of unity are very particular cases of infinite generalized Kumrner extensions. W e present in this section their basic properties.
DEFINITION 13.3.1. We say that an extension E I F is an n-Kummer extensions with few roots of unity if E = F ( B ) for some 0 # B E E * , {-1,l). with gcd(n,e(F))= 1, B n C F and p,(E) 0
PROPOSITION 13.4.3. Any Galois n-bounded G-Cogalois extension is a quasi-Kummer extension. PROOF. Let E I F be a Galois n-bounded G-Cogalois extension. Then E = F ( G ) , with F* 6 G 6 E* and e x p ( G / F * )= n. Since the extension E I F is Galois, it follows that gcd(n,e(F))= 1 and Cn E E by Corollary 5.1.6, so pn(R) c p,,(E). B y Theorem 12.1.12, the G-Cogalois extension E E I F is n-pure, hence p p ( E ) C F for every p E P,. Then p p ( E ) F for any such p. Consequently, the extension E I F is quasicl Kummer.
An immediate consequence of Theorem 13.2.2 is the following result.
THEOREM 13.3.2. Let E I F be an n-Kummer extension with few roots of unity, with 0 # B C_ E * , E = F ( B ) , Brl F , gcd(n,e(F))= 1, and p,(E) { - 1 , l ) . If we denote G = F * ( B ) , then the following statements hold. ( 1 ) The extension E I F is G-Cogalois. ( 2 ) The maps H I+ F ( n G ) and K I+ K" fl (F*"(B n ) ) establish isomorphisms of lattices, inverse to one another, between the lattice of all subgroups H of F * " ( B T 1 )containing F*n and the lattice of all intermediate fields K of E I F . Moreover, any subextension K I F of E I F is an n-Kummer extension with few roots of unity. ( 3 ) If H is any subgroup of F*"( B n ) containing F*", then any set of n G ) / F * is a vector space basis representatives of the group o f ~ ( m n ~ ) o v e r F , a n d [ ~ ( m n ~ ) : ~ ] = 0 I ~ / ~ * ~ ' I .
&(a)
PROPOSITION 13.4.4. Let E I F be a Galois generalized m-Kummer F, E * , gcd(m,e(F))= 1, B " extension, with E = F ( B ) , 0 # B and p,,&(E) F . Then E I F is F*( B)-Cogalois and p,(R) C F , where 1% = exp(F*( B ) / F * ) , i.e., E I F is a classical n-Kummer extension.
(m
PROOF. Let p E Pn Since n l m , we have /L,(E) p,,&(E) F , hence E I F is n-pure. Observe that E I F is G-radical, where G = F * ( B ) . By Theorem 12.1.12, E / F is a G-Cogalois extension. Now, in view of E E . But n I m, hence Proposition 5.1.6, we have gcd(n, e ( F ) )= 1 and <,,
Thus E I F is a classical n-Kummer extension.
288
Chapter 13
13.6 Uibliographical corrlrner~ts to Chapter 13
289
The last result of this section establishes the structure of Abelian bounded G-Cogalois extensions.
PROPOSITION 13.4.5. The following assertions are equivalent for an extension E / Q with E C @. (1) E l Q is an Abelian n-bounded G-Cogalois extension. (2) There exists 0 # B C E* such that B 2 C Q and E = Q ( B ) . (3) There exists 0 # A C_ Q such that E = Q(JZ), where denotes the set { z E @I z2 E A), in other words, E is obtained by adjoining to Q an arbitrary nonempty set of square roots of rational numbers. (4) E / Q is a classical 2-Kummer extension.
PROOF. (1) (2): By Proposition 13.4.3, E l Q is a quasi-Kummer extension. Then E = Q ( B ) for some 0 # B C E*, there exists n E N* with B7& E Q, and E Q for every p E P,,. The liwt condition implies " Q. that n = 1 or n = 2, so E = Q ( B ) with B
P :Q 1. Let F be a field, let n E N* be such that p7,(R) F , and let F be the maximal classical n-Kummer extension of F contained in a fixed algebraically closed overfield f l of F . Prove that
H O ~ , ( G ~ ~ ( F ; % / Fp,(F)) ), 2 F*/F*". 2. With the notation and hypotheses of Exercise 1, prove that G~~(F,$~/F) Z Hom(F*, 3. Show that
= Q(i,
(F)).
a, fi,&,. ..).
cp
" Q is clearly Abelian (2) ==+ (1): Any extension Q ( B ) / Q with B and Q (A)-Cogalois. It is 1-bounded if Q = Q ( B ) , and 2-bounded otherwise.
All other implications are obvious.
4. Prove that the following assertions are equivalent for an extension EIF with F C R and E C_ @. (a) E/F is a Galois n-bounded G-Cogalois extension. (b) E/F is an Abelian n-bounded G-Cogalois extension. F and E = F ( B ) . (c) There exists 0 # B E* such that B' F* such that E = ~(fi), where (d) There exists 0 # A denotes the set { z E @ I z' E A ) , in other words, E is obtained by adjoining t o F an arbitrary nonempty set of square roots of numbers from F . (e) EIF is a classical 2-Kummer extension.
R.EMARK13.4.6. Recall that we have denoted by I F ' the set of all strictly positive prime numbers. Clearly, Q(JF)IQ is an Abelian 2-bounded GCogalois extension, where G = The Kneser group of this extension is Q ( a ) / @ , which is isomorphic to a countably infinite direct sum of copies of the cyclic group Z2 of order 2. On the other hand, the Galois group of this extension is isomorphic t o a countably infinite direct product of copies of 2 2 . Consequently, the Kneser group arid the Galois group of the Abelian G-Cogalois extension Q ( J F ) / Q are rlot isomorphic. This shows that Theorem 5.2.2, saying that the Kneser group and the Galois group of any finite Abelian G-Cogalois extension are isomorphic, is no more valid in the infinite case. The connection between Gal(E/F) and Kne(E/F) for an infinite Abelian G-Cogalois extension E I F , which involves also the Pontryagin dualit.y, will be investigated in Section 15.3. 0
@(a).
Section 13.1. The concept of classical Kummer extension is rather basic and well-known, so it is difficult t o give many precise attributions. A good account of the theory of such extensions can be found e.g., in Bourbaki [40], Karpilovsky [76], and Lang [80]. The idea t o investigate classical Kummer extensions via Finite orland Infinite Cogalois Theory is due t o Albu and Kicolae [19] and Albu and Tena [25]. Classical but longer proofs of Theorem 13.1.2 are provided e.g., in Bourbaki [40, Chap. V, $11, Thkorkme 4, p. 851, Karpilovsky [76, Chap. 7, Theorem 4.4, p. 4121, or Lang [80, Chap. 8, Theorem 141. Section 13.2. The results of this section are taken from Albu and Tena [25], where the notion of infinite generalized Kumrner exterlsion was introduced and investigated.
290
Chapter 13
Section 13.3. The concept of Kummer extension with few roots of unity was introduced and investigated by Albu [3] for finite extensions, and by Albu and Tena [2,5] for infinite extensions. The contents of this section follow the presentation in Albu and 'J'ena [25]. Section 13.4. The notion of (finite) neat presentation is due t o Greither and Harrison [63]. Albu and Nicolae [I91 introduced the more general concept of (finite) generalized neat presentation. The infinite generalized neat presentations were introduced and investigated by Albu and Tena [25]. The results of this section on quasi-Kummer extensions are taken from Albu and Tena [25], where they were called generalized neat presentations.
CHAPTER. 14
INFINITE GALOIS THEORY AND PONTRYAGIN DUALITY

In this chapter we present without proofs some basic facts on Infinite Galois Theory and Pontryagin Duality which will be used in the next chapter. The reader is iusurned t o be familiar with the concept of projective limit and with the basic concepts ancl facts of General Topology. 14.1. Profinite groups and Infinite Galois Theory The main aim of this section is to present without proofs the Fundamental Theorem of the Infinite Galois Theory and its consequences. The Galois group of any infinite Galois extension EIF turns out t o be a profinite group, that is, a topological group which is the projective limit of a projective system of finite groups, each endowed with the discrete topology. Therefore we shall quote some standard properties of profinite groups which will be applied t o Infinite Galois Theory. A good account of profinite groups can be found in Cassels and Friihlich [46, Chapter 51 or Karpilovsky [76, Chapter 6, Section 21.
DEFINITION 14.1.1. A topological group G is said t o be a profinite group if G is a projective limit of a projective system (Gi, p j i ) i , j E ~ of 0 finite groups, each endowed with the discrete topology.
Recall briefly some definitions from General Topology. Let S be a topological space. A fundamental system of neighborhoods of a point x of S is a set U of neighborhoods of x sudi that for every neighborhood V of x there exists U E U with U E V. The space S is said t o be totally disconnected if the connected component of each point of S consists of the point alone. The topological space S is said to be compact if for every family ( A i ) i E ,of open subsets
202
Chapter 1 4
14.1 Profinite groups and Infinite Galois 'l'heory
293
Ai of S with S = S = UiEJ Ai.
Ui,,
Ai there exists a finite subset J of I such that
For other basic terminology from General Topology, the reader is referred t o Bourbaki [38].
PROPOSITION 14.1.2. The following assertions are equivalent for a topological group G. (1) G is a profinite group. (2) G is a Hausdorff, compact group which has a fundamental system of open neighborhoods of the identity element e of G consisting of normal subgroups of G . (3) G is a Hausdorff, compact, totally disconnected group.
PROOF. See Karpilovsky [76, Proposition 2.8, p. 3111.
The topology on Gal(E/F) defined in Proposition 14.1.4 is called the Krull topology; note that it is the discrete topology if and only if the Galois extension E/F is finite. We are going t o show that the topological group G a l ( E / F ) is a profinite group. To do that we will show first that the Krull topology on Gal(E/F) is nothing else than the finite topology. R,ecall that if X and Y are two nonempty sets, then the finite topology on the set yX of all maps from X t o Y, identified with the Cartesian product Y,, with Y, = Y for all x E X , is the product topology on y X , where Y is endowed with the discrete topology. Taking into account the form of open sets in the product topology, we deduce a t once that for an arbitrary f E yX a fundamental system of open neighborhoods of f consists of the sets
nzEx
0
where {xl,. . . ,x , ~ ) ranges over the finite subsets of X. For any Z y X , by the finite topology of Z we will understand the topology on Z induced by the finite topology 011 Y X . If EIF is a Galois extension, then clearly Gal(E/F) E E , hence it makes sense t o consider the finite topology on Gal(E/F).
The prototype of a profinite group is the Galois group r of an infinite Galois extension E / F . The group I' can be endowed with the Km11 topology as follows. Denote by F the set of all intermediate fields K of E/F such that K / F is a finite Galois extension, and for any K E F set rK = Gal(E/K). Notice that for any K , K1, K 2 E F , one has I'KQ I ' , K1K2 E F , and rKIKz rK1 nrK, (see?1.2.9).
LEMMA 14.1.3. Let G be a group, and let Jtf be a nonempty set of normal subgroups of G such that for every N l , N2 E Jtf there exists an No E Jtf with No E Nl n N2. Then G becomes a topological group if one takes Jtf as a fundamental system of open neighborhoods of the identity element e of G . A fundamental system of open neighborhoods i n this topology of any x E G is the set { x N I N E Jtf ). Moreover, this group is Hausdorff if and only if n N E N N = {e).
PROOF.See Bourbaki [38, Chap. 111, $1, no.2, Example].
PROPOSITION 14.1.5. The following statements hold for a Galois ex' . tension E/F with Galois group I (1) The Krull topology and the finite topology on I' coincide. (2) I ' is a Hausdorff, compact, and totally disconnected group. (3) r is a profinite group.
PROOF. (1) Let o E I ' , and let U be a neighborhood of o in the finite topology. Then, with the notation above, there exists a finite subset )(o) U. Set K = F(x1,. . . ,I,). { x l , . . . , x , } of E such that V{,,,...,,,, Since K / F is a finite extension, so also is its normal closure K/F. But
PROPOSITION 14.1.4. Let E/F be a Galois extension. W i t h the notation above, Gal(E/F) is a Hausdorff topological group if one takes the set { I'KI K E F ) as a fundamental system of open neighborhoods of l E .
PROOF. We have seen that the set { rKI K E F ) satisfies the conditions of Lemma 14.1.3. We claim that I'K = {IE). Indeed, let E- I'K and x E E . Set L = F ( x ) , and consider the normal closure oE of E/F of the finite extension L I F . Then 2 E F , hence o E G ~ ~ ( E / E ) , in particular we have o(x) = x. This proves our claim. Kow apply Lemma 0 14.1.3.
nK
nKEF
Thus,
o~al(~1K ) c oGal(EIK) = V{z,,... ,,,J(o) K which shows that for every o E I", every neighborhood of o in the finite topology includes a neighborhood of o in the Krull topology.
294
Chapter 1 4
14.1 Profinite groups arrd Infinite Galois Theory
295
Conversely, let a E I?, and let W be a neighborhood of a in the Krull topology. Then, there exists a finite Galois subextension L I F of EIF such that aGal(E/L) C W. Let {yl, . . . ,y,,) be a basis of the finite dimensional vector space pL. Then L = F ( y l , . . . ,y,,), and so,
as we have already shown. Thus, for every a E I?, every neighborhood of a in the Krull topology includes a neighborhood of a in the finite topology. This proves (1).
(2) I? is a Hausdorff topological group by Proposition 14.1.4. For every
E E set E, = E , and let a, : E, --+ E, denote the canonical a:-th projection. For any z E E , the set a,(I?) = { a ( x ) I a E I? ) is the set
3:
Galois extension EIF and the lattice Intermediate(E1F) of all intermediate fields of E I F , provided by the Fundamental Theorem of Finite Galois Theory, no longer remains valid if the Galois extension E/F is infinite. However, if we equip the Galois group I? with the Krull topology, then the bijective Galois correspondence is restored when we restrict ourselves only to closed subgroups. Next we will present without proofs the basic facts of the Infinite Galois Theory.
THEOREM 14.1.6. ( T H E FUNDAMENTAL THEOREM OF INFINITE GA-
rms THEORY).The following statements hold for an arbitrary Galois extension EIF with Galois group I?. (1) The maps
a : Intermediate ( E I F )
nu,,
of all conjugates of x over F , hence it is finite since x is algebraic over F . Thus, every 7 ~ ,(I?) is a compact space, hence so also is a, (I?). We claim that I? is a closed subset of E" = E,. To do that, let f E EE be an element in the topological closure of I? in E". By (I), the Krull topology on I? is nothing else than the finite topology, hence, for {, + ,,,, 1 (f) # 0 , i.e., there exists every fixed x , y E E we have I? n V a E I? such that
n,, n,,,(
-+
Closed Subgroups (I?),
a ( K ) = Gal(E/K)
and
8 : Closed Subgroups (I?) -+
Intermediate ( E I F ) ,
It follows that f ( x + y) = f ( x ) f(y) and f(xy) = f(x)f(y). A similar argument shows that f (a) = a for every a E F. Thus f : E -+ E is an F-morphism of the field E , which is an F-automorphism of E , i.e., f E I? since EIF is an algebraic extension. This proves that I? is a closed subset a, (I?), hence of E". Then I? is a closed subset in the compact space it is also compact. Every a E I? has as a fundamental system of open neighborhoods the set { a G a l ( E / K ) I K E 3). Since every open subgroup of a topological group is closed, it follows that every member of this set is also closed. This implies that the connected component of any a is contained in n K E F ~ G a l ( E / K = ) {a), SO it is reduced t o {a); in other words, I? is totally disconnectecl.
n,,,
b(A) = Fix(A) establish anti-isomorphisms of lattices, inverse to one another, between the lattice Intermediate ( E I F ) of all intermediate fields of the extension EIF and the lattice Closed Subgroups (I?) of all closed subgroups of the group I? endowed with the Krull topology. (2) If K E Intermediate ( E I F ) , then K I F is a finite extension if and only if Gal(E/K) is an open subgroup of I?, and i n this case one has ( I ? : Gal(E/K)) = [ K : F ] . (3) If K E Intermediate(E/F), then K I F is a Galois extension if and only if Gal(E/K) a Gal(E/F), and i n this case, the restriction map Gal(E/F) --+ Gal(K/F) yields an isomorphism of topological groups Gal(E/F)/Gal(E/K) % Gal(K1F).
PROOF. See Bourbaki [40, Thkorkme 4, A V.641 or Karpilovsky [76, Theorem 6.2, p. 3431. 0 COROLLARY 14.1.7. Let EIF and K I F be subextensions of an extension L I F . If E / ( E n K ) is a Galois extension, then E K I K is also a Galois extension, and the restriction map: Gal(EK1K) --+ Gal(E/(E n K ) ) , a
I+
(3) follows from (2) and Proposition 14.1.2. A more accurate description of the realization of the Galois group of an infinite Galois extension as a projective limit of finite groups is given in Exercise 3. 0 -4s Exercise 4 shows, the bijective Galois correspondence between the lattice Subgroups(I') of all subgroups of the Galois group I? of any finite
is an isomorphism of topological groups.
290
Chapter 14
14.2 Character group and I'or~tryagirl Iluality
297
COROLLARY 14.1.8. Let E1/F and E2/F be Galois extensions, and assume that E l , E2 are subfields of some other field. Then ElE2/F is a Galois extension, and the map Gal(El&/J')
(3) If G is a profinite group, then Ch(G) is a torsion Abelian group, and

Ch (G) = { x E Hom(G, ,u(C)) ( Ker (x) is open ) = Hom,(G,p(C)). (4) If G is a finite discrete Abelian group, then Ch(G) Z G. (5) If G is a locally compact Abelian group, then so is Ch(G). PROOF. (1) If G is a torsion group, then for every x E G there exists an n, E W such that xn= = e, where e is the identity element of G. For any x E 8 we have 1 = x (e) = x (xr'=) = (X(x))"", hence ~ ( x E) p(C). (2) If G is an n-bounded group, then x" = e for every x E G, so, as in the proof of ( I ) , we have ( ~ ( x ) ) = " 1 for every x E and every x E G, i.e.7 x ( 4 &l(C).
(3) See Karpilovsky [76, Corollary 5.3, p. 4181.
--+ Gal(&/F)
x Gal(EL/F), a
I+
( ~ I E,,u ~ E ~ ) ,
is a monomorphism of topological groups, which is an isomorphism when El n E2 = F.
14.2. Character group and Pontryagin Dualitj In this section we present some basic facts on characters of locally compact Abelian groups and Pontryagin Duality. If G and H are topological groups, then by Hom ,(G, H ) we denote the set of all continuous group morphisms from G t o H, while Hom(G, H ) denotes the set of all group morphisms from G t o H . By a' we denote the unit circle of the complex plane, that is, If = ( z I x E C, Izl = 1). Observe that J ' l' is a subgroup of the multiplicative group C . If we consider on If the induced topology of the usual topology of C, then it becomes a topological group. For any topological group G we denote by Closed subgroups(G) the lattice of all closed subgroups of G, and by Ch (G) or by 8 the character group of G, that is, the group of all continuous morphisms of G into the unit circle T: Ch (G) = (? = Horn ,(G, T). By 1 we shall denote the 1-character of G, that is, l ( x ) = 1 for all x E G.
(4) See Hewitt and Ross [72, 23.27 (d)].

E
> 0, let
(5) For any nonempty compact subset C of G and any real number defined by U(C, E) be the subset of
If one takes the collection of all such U(C,&) as a fundamental system of neighborhoods of the identity element 1 E (?,then (? becomes a locally compact Abelian group by Hewitt and Ross [72, Theorem 23.151. 0 Throughout this section A will denote a fixed multiplicative Abelian group with identity elernent e. By Lemrna 14.2.1 (5), for any locally compact Abelian group A, the character group Ch (A) = A^ of A is again a locally compact Abelian group, so it makes sense t o consider the character
A
LEMMA 14.2.1. The following assertions hold for a topological group

G.
( 1 ) If G is a torsion group, then
group Ch (Ch (A)) the canonical map
=A^
of
2, called the second
character group of A, and
Ch(G) = Hom ,(G, ,u(C)). (2) If G is an n-bounded group, then The well-known Duality Theorem, due to Pontryagin, establishes the following fundamental result.
298
Chapter 1 4
1 4 . 2 C h a r a c t e r group a n d I'orltryagin I h a l i t y
A
299
THEOREM 14.2.2. (THE PONTRYACIN DUALITY).The following assertions hold for any locally compact Abelian group A.
(1) The canonical map wa : A --+ A^ is a topological isomorphism. (2) The assignment A ct A^ establishes a n auto-duality of the category L C A of all locally compact Abelian groups. (3) A is a compact group if and only if is a discrete group. ( 4 ) A is a discrete group if and only if is a compact group. (5) If A is a profinite group then is a discrete torsion group.
-.
such that a* = wA(a),hence a*($) = w~ (a)(?/))= ~ b ( a )for all $ E A. In particular, we have a*(x) = ~ ( a # ) 1. On the other hand, 1 = a*(li',)= +(a) for every :!I E Y, hence a E Y ' - . Since x E Y' l we deduce that ~ ( b= ) 1 for all b E Y l , hence ~ ( a = ) 1, which is a contradiction. This proves (5). 0
A^ A^
A^
PROPOSITION 14.2.4. Let A be a locally compact Abelian group. Then, the maps: Closed Subgroups (A) + Closed Subgroups (A^), X I+ X',
Closed Subgroups (2)t Closed Subgroups (A), Y are anti-ison~orphisnis of lattices, inverse to one another.
I+
PROOF. See Hewitt and Ross [72, Theorem 24.81 for (I), and Hewitt A and R.oss [72, Theorem 23.171 for (3) and (4). Since the assignment A is functorial, (2) follows from (1). Finally, (5) follows from Proposition 14.1.2, Lemma 14.2.1, and (3).
h
YL,
PROOF. For si~nplicity, set

A = Closed Subgroups ( A ) and
For any notation:
# X 5 A and
# Y
A^
we shall use the following
A^ = Closed
Subgroups ( 2 ) .
For every Z in d or A^ one has Z = ZLi in other words, the order-reversing maps
by Lemma 14.2.3 (4) and ( 5 ) ,
are poset anti-isomorphisms, and consequently, lattice anti-isomorphisms, inverse to one another. Ci
LEMMA 14.2.3. The following statements hold for any locally compact Abelian group A, any closed subgroup X of A, and any closed subgroup Y of 2. (1) X I is a closed subgroup of 2. (2) YL is a closed subgroup of A. (3) If X # A and x E A \ X , then there exists $ E X I such that +(x) # 1. (4) X = X I L . (5) Y = YLL.
PROOF. For (1) see Hewitt and R.oss [72, Remarks 23.24 (c)]. Similarly, (2) is easily verified. For (3) see Hewitt arid Ross [72, Corollary 23.261, and for (4) see Hewitt and R.oss [72, Theorem 24.101. We are going to prove (5). Assume that Y # YLL. This means that there exists E YLL \ Y. By (3), applied to the subgroup Y of and t o
COROILARY 14.2.5. The following statements hold for a topological Abelian group A. (1) If A is a compact group, then the map Closed Subgroups (A) ---+ Subgroups (A), X
I+
x',
is a n anti-isomorphism of lattices. (2) If A is a discrete group, then the map

Subgroups (A)
--+ Closed Subgroups (A^),
rt
x',
is a n anti-isomorphism of lattices.
PROOF. By Theorem 14.2.2 (3) and (4), if A is a discrete (resp. cornpact) group, then ,4 is a compact (resp. discrete) group. Since
Closed Subgroups (D) = Subgroups (D) for any discrete topological group D , the results follow a t once from Propo0 sition 14.2.4.
A^
the character X, we deduce that there exists a* 2 such that a*(q5) = 1 for all @ E Y and a*(x) # 1. By Theorem 14.2.2 ( I ) , there exists a E A
300
Chapter 1 4
14.3 ICxercises to Chapter 1 4
301
1. Show that the following statements hold for a topological group G. (a) Every open subgroup of G is closed. (b) Every closed subgroup of G of finite index is open. (c) If G is compact, then a subgroup of G is open if and only if it is closed and of finite index.
2. Let G be a compact topological group, and let (Ni)iEl be a family of closed norrnal subgroups of finite index satisfying the following two conditions. (4 hTi= {e). (b) For every i l , i2 E I there exists io E I such that Nio C Ni, n Ni,. If we set i ,< j for A T i ATj and cpij : GINj + GINi for the canonical projection, then prove that (GINi, pij)i,j E I is a projective system of topological groups, and there exists an isomorphism of topologicid groups G r li~n GIN,.
ni,,
>
6. Let { I ) # S be a multiplicative subset of K (this means that 1 E S and ntn E S for any nt, n E S ) . Then S becomes a directed set if one orders S by the relation of divisibility. For every nurnbers m 1 n in S let p,, : Z,, + Z, denote the canonical projection. Prove the following statements. cp,n7,)7n,,lis a projective system of rings. Denote by (a) ( P ( S ) the corresponding projective limit, which is a compact , with the discrete topological ring if one endows each ring Z topology. If S = K then P ( S ) is denoted by and is called the Priifer ring. If S = { prLI n E N), where p > 0 is a prime number, then P ( S ) is denoted by Z , and is called the ring of p-adic integers. (b) The underlying additive group of Z (resp. Z p ) is i m Abelian profinite group, having as open subgroups precisely the cyclic subgroups n z , n t PI' (resp. p711Zp, ni E N). (c) There exist canonical group isomorphisms:
Zi
3. Let E/F be a Galois extension, and denote bj. . F the poset (ordered by inclusion) of all intermediate fields K of EIF such that K / F is a finite Galois extensions. (a) For any K L in F denote by the restriction morphism. Show that (Gal(K/F), pKI,)K,LE3 is a projective system of discrete topological groups. (b) Prove that there exists an isomorphis~nof topological groups G a l ( E / F ) 2 lirn Gal(K/F).
t-
for any n E N* and any nt E N. (d) There exists a canonical isomorphism of topological rings:
7. Prove that for any profinite group G and for any given g E G there , Z -+ G such that exists a unique continuous group morphism c ~ :
4 1 ) = 9.
KEF
4. Let E be an algebraic closure of the field F = IFp of p elements, p prime, let T = Gwl(E/F), let cp E T be the Frobenius auto E, and let 4 = (p). Show that morphism of E , p(x) = z\ there exists no K E Interrnediii.te(E/F) such that 4 = Gal(E/K). (Hint: See Kwpilovsky (76, Example 5.5, p. 3381.)
8. (Serre [97, p. 1981). A field F is said to be quasi-finite if it is perfect '* 2. Show that any finite field and the field of formal and G ~ ( F / F ) power series K ( ( T ) ) in the indeterminate T over any algebraically closed field K of characteristic 0 are quasi-finite fields.
9. Show that if EIF is an infinite Galois extension with Galois group I ' ,then the cardinal numbers [ E : F ] and 1 I' 1 are not necessarily equal.
10. Let ( E i / F ) ,<i<,, be il finite family of Galois extensions, and assume that all E l , . . . ,E,, are subfielcls of some other field. If
5. Let E/F be a Galois extension with Galois group r, let A be any be the topological closure of A in T. subgroup of T, and let Show that Gal(EIFix(4)) = &.
Chapter 1 4
for every i , 1 i n - 1, then show that there exists an isornorphism of topological groups
< <

Section 14.1. This section contains some well-known classical facts on
11. Let E l , E2 be intermediate fields of a Galois extension E/F, and let r = G a l ( E / F ) , rl = G a l ( E l / F ) , r2= Gal(E2/F). Prove that the following assertions a.re equivalent. (a) The group r is the internal direct products of its subgroups rl and r 2 . (b) El I F and E2/F are both Galois extensions, El n E2 = F , and E = E l E 2 .
12. (Waterhouse [ I l l ] ) . Prove that every profinite group is isomorphic to the Galois group of some field extension.
Infinite Galois Theory and profinite groups. Our rnain sources of inspiration for this section were Bourbaki [40] and Karpilovsky [76]. An abstract Cogalois Theory, applicable t o an arbitrary profinite group r acting continuously on a discrete subgroup of the additive group Q/Z, is developed in ,41bu and Basarab [16]. This theory is somewhat dual to the abstract Galois Theory which plays a major role in the general Class Field Theory as presented in n'eukirch [85].
Section 14.2. A good account of Pontryagin Duality, only sketched in this section, can be found in Hewitt and R.oss [72]. We also used the monograph of Karpilovsky [76] for our presentation. Proposition 14.2.4 is certainly known, but we could not locate it explicitly in the literature.
13. Prove that Ch(2) Z Q/Z and Ch(2,) lian group A4. (a) The evaluation map on
E t,(Q/Z)
for any p E
IP.
14. Show that the following staternelits hold for a locally compact Abe-
is bimultiplicative. (b) For any 0 # X A and any
# Y 2 A^ one has
(c) The evaluation map on A is nonsingular, that is, A" = (1) and 2~= {e). 15. Give an example of a compact Abelian group A which is not profinite, but for which A is a t,orsion group. This shows that the converse of Theorern 14.2.2 (5) is not true in general. (Efint: See Hewitt and R.oss [72, 7.18 (ti) and Theorem 24.261.)
A
CHAPTER 15
INFINITE GALOIS G-COGALOIS EXTENSIONS

The aim of this chapter is two-fold: first, to generalize the main results of Chapter 6 from finite t o infinite Galois extensions, and, second, t o investigate infinite Abelian G-Cogalois extensions. In Section 15.1 we study infinite Galois extensions EIF by means of continuous crossed l~omornorphisms of G a l ( E / F ) with coefficients in the group p ( E ) of all roots of unity of E. We show that a considerable part of the results of Chapter 6 can be naturally generalized t o infinite extensions. Next, we are interested in finding a relationship between the Abelian groups K n e ( E / F ) and G a l ( E / F ) associated with an arbitrary Abelian GCogalois extension E I F . When the extension E/F is finite, then, by Theorem 5.2.2, these two groups are isomorphic, but not in a canonical way. We show that in the infinite case, K n e ( E / F ) is isomorphic t o the group Ch(Gal( E I F ) ) of characters of the profinite group Gal(E1F). To do that we cannot proceed, as in the finite case; by incluction on the degree of the extension E I F . As we sketched in Remark 5.2.4, for any Abelian G-Cogalois extension EIF, finite or not, there exists a canonical lattice isomorphism between the lattices Subgroups ( K n e ( E / F ) ) and Subgroups (Ch(Gal(E I F ) ) ) . So, it is natural t o ask whether or not such a lattice isomorphism yields a group isomorphism between K n e ( E / F ) and Ch(Gal(E1F)). Thus, we arrive a t the following question: given two groups A and B , when does a lattice isomorphism y : Subgroups ( A ) -+ Subgroups(B) give rise a group isomorphism f : A - + B ? The groups A and B are called lattice-isomorphic if there exists a lattice isomorphism between their subgroup httices Subgroups (,A) and Subgroups(B). With this terminology, the problem we just stated c i ~ n be reformula.ted briefly as follows: when are two lattice-isomorphic groups isomorphic'? In general,
13.1 'I'lle ir~firliteK ~ l e s e r group via crossed horr~o~r~orphisrr~s
307
lattice-isomorphic groups are not isornorpliic, as a classical example discovered in 1928 by -4. R.ottlaender shows (see Baer 1271). However, if some restrictive conditions on one or both groups A arid B are imposed, then any lattice-isomorphism between ,4 and B produces a group isomorphism between them. By chance, such conditions are satisfied in our case, a i d so, the canonical lattice isomorphism between Subgroups (Kne ( E I F ) ) and Subgroups (Ch (Gal ( E I F ) ) ) yields 1-1 group isomorphism between the groups Kne ( E I F ) and Ch (Gal (ElF ) ) . In section 15.2 we present the basic notions and facts concerning latticeisomorphic groups. Since many of these facts are quite technical, we shall a.void most of their proofs. Then, we apply in Section 15.3 these general results on lattice-isornorphic groups t o our specific Galois-Cogalois case. Observe that this approach of the infinite case provides an alternative proof of Theorem 5.2.2 via lr1,ttice-isornorpliisrnsof groups, which is more natural than that given in Section 5.2. Intleed, for any finite extension E I F , the finite group G a l ( E / F ) is iso~norphic,but not in a canonical way, t o its character group Ch (Gal( E I F ) ) .
index, that is, of subgroups of the form Gal ( E I N ) , with N I F is a finite normal subextensiori of E I F . As in the finite case, let M 6 E* be such that a(&!) C A/l for every aE I ' . A continuous crossed homomorphism or a continuous 1-cocycle of I ' with coefficients in M is a continuous function f E Z1(I', M ) , where M is endowed with the discrete topology. The set of all continuous crossed homomorphisms of I' with coefficients in M is a subgroup of Z1 (I', M), and will be denoted in the sequel by Z: (I?, M ) . We claim that for every a: E M, the 1-coboundary fa : I ' + M defined rEI ' , is a. continuous map. To see as in Section 6.1 by f,(o) = a ( a ) a - ' , c this, observe that a map h : P -t D, where P is a profinite group and D is a discrete set is continuous if and only if 11 is locally constant, t h a t is, there exists a normal subgroup H in P of finite index such t h a t h factorizes through the canonical surjection map P + P I H . Let a E M , and denote by AT/F the normal closure of F ( a ) / F , witli N E . Then N / F is a finite Galois extension, and for all cr E Gal ( E I F ) and v E Gal ( E I N ) one has
15.1. The infinite Kneser group via crossed homomorphisms
hence f,, factorizes through the canonicrd map Gal ( E I F )

---t
The main purpose of this section is t o generalize to infinite Galois extensions a series of results from Chapter 6 concerning the investigation of finite Galois extensions via crossed homomorphisms. As we know from Section 14.1, the Galois group of a n infinite Galois extension is a totally disconnected compact topological group. In order t o obtain these generalizations, one replaces ntutatis-mutandis the (discrete) crossed hornomorphisms by the continuous ones. We also characterize infinite radical extensions, Kneser extensions, and G-Cogalois extensions via crossed ho~no~norpliisms, generalizing the corresponding results established in Section 6.1 for finite extensions. In Section 6.1 we reviewed same basic facts on finite Galois Cohomology. In particular we stated t,he (finite) Hilbert's Theorem 90. \\'hen
Gal ( E / F ) / G a l ( E I N ) .
EIF
This proves our claim. M ) = { f, I cr E M ) of all 1-coboundaries Consequently, the set B' (I', of I' with coefficients in M coincides with the set B: (I', M ) of all continuous 1-coboundaries of I' witli coefficients in M . The quotient group M)/B:(I', M) is called the first continuous cohomology group of I ' Z: (I', with coefficients in M , and is denoted by Hi (T, M). Note t h a t when EIF is a finite Galois extension, then Z: (I', M) = Z1(I', M ) , hence HI (I', M ) = HI (I', M ) . The continuous (or infinite) Hilbert's Theorem 90 asserts that if EIF is an arbitrary Galois extension, E * ) = 1 (see e.g., Cassels and Friihlich [46, finite or infinite, then H: (I', Proposition 2.2, Chapter V] or Serre [96. Proposition 1, Chapitre 21). For an arbitrary extension EIF we have considered in Section 6.1 the map
is an infinite field extension, then a n infinite Hilbert's Theorem 90 still ' of E / F is a holds. As we have seen in Section 14.1, the Galois group I profinite group, or equivalently, a Hausdorff, compact, and totally clisconof nected topological group. A fundamental system of open ~ieigliborhoods the identity element I]:. of I ' consists of normal subgroups of I ' of finite
f : Gal ( E I F ) x Cog ( E I F ) -+ p ( E ) ,
15.1 'I'tle infinite K ~ l e s e r group via crossed tlor~lorrlorphisrrls
and for every integer n 2 1, its restriction to Cog,,(E/F)

fn :
PROOF. For any n 2 1, the restriction of the rnorphism
Gal ( E I F ) X Cog,,(E/F)
---)
PI, ( 0
w : Cog ( E I F ) -+ 2: (Gal ( E I F ) ,p ( E ) )
t o Cog,,(E/F) is +,. Since T = Gal ( E I F ) is a profinite group, every hE 2 : (I', p ( E ) ) is locally constant, and so,
For every fixed a E Gal ( E I F ) , the partial map f (a, -) is clearly multiplicative on Cog ( E I F ) , and for every fixed G E Cog(E/F), the partial map f (-, Zi) is precisely the 1-coboundary f, E 2' (Gal ( E I F ) , p ( E ) ) , so f and fn induce morphisms of groups
b !
: cog ( E I F )
+ 2' (Gal ( E I F ) ,p ( E ) ) ,
= f (0, q,
But $,, is bijective for all n E W by Lemma 15.1.1, hence too.
7,b
is bijective
The Galois group Gal ( E I F ) = I ' is a profinite group, in particular it is a topological group with respect to its Krull topology. The morphisms 7 1 ' and I/),, have images in the corresponding groups of continuous crossed 2: (T, p ( E ) ) . Consehomomorphisms since $(Zi) = f, E B1(T, p ( E ) ) quently. we can assume that the carmlical maps 7fi and TI', are defined as follows: li/ : C o g ( E / F ) and
COROTJARY 15.1.3. If EIF is a Galois extension with Galois group I ' , then the map P:{HIF* < H < T ( E I F ) ) + { U ~ U < ~ E ( ~ , P ( E ) ) ) , P ( W = { fa E 2: (T, I@)) 1 a E H ), is a lattice isomorphism, which induces a canonical lattice isomorphism Subgroups(Cog ( E I F ) )
E
Subgroups (2: (I?, p ( E ) ) ) .
+2 ; ( G a l ( E / F ) , p ( E ) ) ,
Ij:(i?)
= fa,
For every cyclic subgroup C of 2 : (T, p ( E ) ) there exists a E T ( E / F ) such that (p(F*( a ) ) = (f,) = C. Moreover, H I F * 2 y ( H ) for every H 0 with F* H T ( E / F ) .
< <
LEMMA 15.1.1. Let EIF be an arbitrary Galois extension, and let n E
N*. Then, the morphism

defined above is an isomorphism of groups.
PROOF.Denote by I ' the group Gal ( E I F ) , and let a E T,,(E/F). As in the proof of Lemma 6.1.1 one shows that Q,, is a monomorphism. We are going to prove that &, is surjective. Let h E 2: (I?, p,,(E)). Since clearly h E 2: (I', E * ) , by the continuous Hilbert's Theorem 90, there exists a E E* such that h = f,,. Now continue as in the proof of Lemma 0 6.1.1 to deduce that h = $,, (G), with & E Cog,, ( E I F ) .
For an arbitrary G-radical extension E/F we shal! use the following notation PGW:= CL.L(E).
7nE OG/ .P*
Recall that
OGIF* := { ord(2) I
X E
G ).
THEOREM 15.1.2. For any Galois extension E I F , the map G establishes a group isomorphism
Cog ( E I F ) r 2,' (Gal ( E / F ) , p(E)).
f,
Since OclF* is a directed set with respect to the divisibility relation by Lemma 1.4.3 (2), p c ( E ) is a subgroup of the group p ( E ) . Also, recall that by PG we have denoted the set OC;lFLn P. Kote that p G ( E ) = p,, ( E ) whenever EIF is n-bounded, that is, whenD , , . When the Gever G / F * has finite exponent n , since then O C ; / p= I radical extension E/F is unbounded, the inclusion pc(E) C p ( E ) may be strict. Let E / F be an arbitrary G-radical extension, and let a E G. If m = ortl(s), then rn E OGII;* and
15.1 ?'he infinite Kneser group via crossed t~ornorriorphisn~s
31 1
hence f,(a)) E p,,,(E) C pc;(E) for a11 a E G a l ( E / F ) , i.e., Im(f,) E PG(E). We deduce that whenever EIF is a G-radical Galois extension, then the group isomorphism
$J : Cog ( E I F ) -+ 2 : (Gal ( E I F ) ,p ( E ) )
considered a t the beginning of this section yields by restriction the map g : Gal ( E I F ) x Kne ( E I F ) -+ p G ( E ) , g ( a , 2 ) = f,(a) = a ( a ) . a - ' . For every A
induces by restriction to G I F * a monornorphism

'
< Gal ( E I F ) and
< Kne ( E I F ) let,
qG : G I F * --+
Z:(G~(E/F),~~(E)).
The next result shows that if additionally the extension EIF is G-Cogalois, then the monornorphism djc; is also surjective, in other words, it is a group isomorphism.
aT= { c ~ K n e ( E / F ) I g ( o , c = ) 1, V ~ A}, E wT = { a E G a l ( E / F ) Ig(a,c) = 1, Vc E W}.

PROPOSITION 15.1.6. For any Galois G-Cogalois extension E I F , the assignments (-)T define mutually inverse anti-isomorphisms between the lattices Closed subgroups (Gal ( E lF ) ) and Subgroups (Kne(E/ F ) ) .
PROOF. The proof is literally the same as that of Proposition 6.1.8 if we replace the words "subgroup of Gal(E/F)" with "closed subgroups of Gal(E/F)" , and therefore is omitted. 0 Let EIF be a Galois extension with Galois group r . Then, by Theorem 15.1.2, there exists a canonical isomorphism Cog ( E j F ) r 2 : (I?, p ( E ) ) , hence the canonical map
2 I+
THEOREM 15.1.4. For any Galois G-Cogalois extension E I F , the map f, yields a group isomorphisrr~
Kne ( E I F )
" 2: (Gal ( E I F ) ,pc(E)).
PROOF.Denote by r the Galois group of E I F . Since r is a profinite group, every f E 2 :( r , p c ( E ) ) is locally constant, and so, taking into account that lcm (s,t) E OaIb,* for all s , t t (?G/b.. by Lemma 1.4.3 (2), we deduce that 2 : ( LP c(E)) =
2;(Lprr,(E)).
T ~ L E O ~ / ~ *
f : Gal ( E I F ) x Cog ( E I F ) -+ p ( E ) ,
f (a, 2 ) = a(a)-cx-',
considered above produces, by replacing Cog ( E I F ) with its isomorphic : (I?,p ( E ) ) , precisely the evaluation map copy 2
Let h E 2 : ( r , r ~ ~ ( E )Then, ). there exists m E 0G/,4* ' such that h E 2: (I?, p m ( E ) ) . But, the map q,,, : Cog,,(E/F) + 2; ( r , p,,(E)) is an isomorphism by Lemma 15.1.1, hence h = +,,,( 2 ) for some a E T,,, (ElF ) . Since clearly P,, E PC, Lemma 12.1.7 (1) irnplies that a E G. Thus 2 E G I F * = K n e ( E / F ) , and so h = $,,(2) = $y,!(Z), which proves that 0 Gc; is a surjective map, and we are done. COROI,I,ARY 15.1.5. Let EIF be a Galois n-bounded G-Cogalois extension. Then
(
For any A
-1
-) :r x 2 : ( r , l-4~))
A E ) , (0, h ) = h b ) .
< r, U < ZA(I',p(E)), and x E Z i ( r , p ( E ) ) we shall denote

A' U'
= { h ~ ~ ; ( r , p ( ~ ) ) I ( a=, ~ h ,) V U E A } ,
= =
{ a r l ( o , h ) = l , VREU}, {aEr'l(CJ,x)=l}.
1 (Gal ( E I F ) , p,, ( E ) ) . Kne ( E I F ) r 2

PROOF. We have already noticed that pG ( E ) = p,, ( E ) whenever EIF 0 is n-bounded. Kow apply Theorem 15.1.4. For any Galois G-Cogalois extension E I F , the map
'X
One verifies easily that A L
< 2: ( r , p ( E ) ) , U' < r , and
= (x)'.
f : Gal ( E I F ) x Cog ( E I F ) --+ p ( E ) ,
The next result, characterizing radical subextensions of a given Galois extension EIF by means of subgroups of 2: (Gal ( E I F ) ,p ( E ) ) , is the infinite variant of Theorem 6.2.1.
31'2
Chapter I 5
15.1 T h e infinite Kneser group via crossed hornorr~orphisn~s
31 3
THEOREM 15.1.7. Let E/F be a Galois extension with Galois group I?, und let K be a n intermediate field of the extension E I F . T h e n K I F is a radical extension (resp. a simple radical extension) if and only if there exists 2: (I',p(E))) such that Gal ( E I K ) = UL U 2: (I?, p ( E ) ) (resp. x (resp. Gal ( E I K ) = X I ) .
<
<
Galois extension, which is not necessarily finite, and L I F is any extension with L n E = F , and such that E and L are subfields of some other field, then according t o Corollary 14.1.7. the map Gal ( E L I L ) 7 Gal ( E I F ) , a I+a~b;, is an isomorphism of topological groups. Using this map, it follows as in Section 6.2 that there exists an injective map p : Radical ( E I F ) -+ h d i c a l (ELIL),
PROOF. The proof is the same as that of Theorem 6.2.1 except that we replace Z1(I?, p ( E ) ) ) with 2: (I?, p ( E ) ) ) and apply the infinite variant 0 of the Fundamental Theorem of Galois Theory. The next result is the infinite variant of Corollary 6.2.3. COROLI~ARY 15.1.8. Let EIF be a Galois extension with Galois group I?, and let K I F be a G-radical subextension of EIF. For every H with F* H G set Ufl = { fa I a E H ) 2: (I?, p ( E ) ) . Then, the following assertions hold. (1) The extension K / F is G - K n e s e r if and only if (I': U;) = lUrll for every H with F* H G and H I F * finite. (2) T h e extension K I F is G-Cogalois if and only if it is G - K n e s e r and the '?erpendicularn m a p V I-+ V L yields a bijection, or equivalently, a n anti-isomorphism of lattices
< <
<
The next result is the infinite variant of Theorem 6.2.4.
< <
{ V I v UG) ---+ { A / c ' & A r, 4 closed subgroup of I ' ) . PROOF. (1) By Proposition 11.1.2, the extension K / F is G-Kneser if and only if for every H with F* H 6 G and H I F * finite, the finite extension F ( H ) / F is H-Kneser, i.e., aiF ( H ) : F j = IH/F*I. If we set All = Gal ( E / F ( H ) ) , then [ F ( H ) : F ] = (I' : 41,) by Galois Theory, and H I F * E UII by Corollary 15.1.3, so IH/F*I = IU1lI. On the other hand, AH = U& according to Theorem 15.1.7. Summing up, we obtain
<
< <
THEOREM 15.1.9. Let E/F be a Galois extension with Galois group I?, and let L I F - be a n arbitrary field extension such that E n L = F . If E and L are subfields of some other field, and p ( E L ) = p ( E ) , then the following assertions hold. (1) GL* 9E* = G for every G with F* G T ( E / F ) . (2) G1 = (GI n E*)L* for every G I with L* GI T ( E L / L ) . (3) T h e m a p
< < < <
<
p : Radical ( E / F )
-+ Radical ( E L I L ) ,
F ( G ) / F I--+ L(GL*)/L, F*
< G < T(E/F),
is bijective, and the m a p

Radical ( E L IL) L(Gl)/L is its inverse.
I+
---+ Radical ( E lF ) ,
F ( G I n E * ) / F , L*
(2) The proof can be achieved by replacing in the proof of Corollary 6.2.3 (2) the words "subgroup of Gal(E1F)" with d o s e d subgroups of Gal(E/F)", and by using the Fundamental Theorem of Infinite Galois Theory instead of the Fundamental Theorem of Finite Galois Theory. 0 Next, we generalize frorn finite t o infinite extensions the results of Section 6.2 concerning the transfer under the base field change of the property of a Galois extension being radical, Kneser, or G-Cogalois. For any extension E/F we have clenotecl in Section 6.2 by R.adical(E/F) the set of all subextensions K / F of E/F which are radical. If EIF is a
< GI < T ( E L / L ) ,
PROOF. The proof is literally the same as that of Theorem 6.2.4, and 0 therefore is left to the reader. R.EMARK 15.1.10. The restriction of the bijection p in Theorem 15.1.9 t o Kneser or t o strongly Kneser subextensions of EIF and E L I L is still bijective by Exercise 4. 0
Chapter 15
16.2 1,attice-isorrlorpt~ic groups
315
15.2. Lattice-isomorphic g r o u p s
The aim of this section is two-fold: first, to provide the terminology and notation coricerning lattice-isomorphisms of groups, and second, t o establish a basic result on lattice-isomorphic torsion Abelian groups, which will be needed in the next section. The reader is assumed to be familiar with the main concepts and facts of Lattice Theory, as exposed e.g., in the monographs of Birkhoff [36], Crawley and Dilworth 11491, or Griitxer [62]. The letters A, B , C, D, G, H, . . . will be used to denote groups. The group morphisms will be denoted by f , g, h, . . . . For any group A, the lattice Subgroups(A) of all subgroups of A will be denoted briefly by IL(A). The index of a subgroup B of a group A, which is either a positive integer or oo,will be denoted as usually by (A : B). We shall denote by G r the category of all groups, by P o s the category of all posets, and by L a t the category of all lattices. The morphisms in the category P o s are the order-preserving (i.e., increasing) maps. Recall that if L and L' are lattices, then a rnorphisrn in L a t frorn L to L' is a map cr : L + L' satisfying the following two conditions: cr(x V y) = a(.)
V
DEFINITION 15.2.1. A lattice-isomorphism from a group A t o a group B is a n y isomorphism p E Isorn Lat (IL(A),lL(B)). T h e lattice-isomorphism p is said t o be induced by a group isomorphism if there exists a n isomorphism f : A -+ B such that p = k ( f ) , and in that case, cp is said t o be induced by f . T h e groups A and B are called lattice-isomorphic if Isom &(A), IL(B)) # 0 , and we denote this situation by A B. 0 Kote that if A and B are isomorphic groups, then, as usually, we shall denote this situation by A B . Clearly, if A, B E G r and A r B , then A ZR, B, but not conversely, as a classical example discovered in 1928 by A. Rottlaender shows (see Baer [27]). Therefore, the following natural question arises:
Given a class X of groups, what kind of conditions (C) o n latticeisomorphisms of groups should be imposed such that for every A E X and for every B E G r , every lattice-isomorphism A El, B satisfying the conditions (C) implies that A r B ?
The next definitions present three conditions for lattice-isomorphisms of groups, which will be involved in answering the question above for X , the c1a.s~ of all Abelian torsion groups (see Theorern 15.2.7). DFFIYITIONS 15.2.2. A lattice-isomorphism p E Isom &(A), lL(B)) between the groups A and B is said t o be index-preserving (resp. strictly index-preserving) if ( C : D ) = (cp(C) : for every cyclic subgroup (resp. subgroup) C of A and for every D 6 C . T h e lattice-isomorphism cp i s said t o be normal if p ( N ) a B for every IV a ,4. 0 The proof of the main result of this section is essentially based on the next four lemmas, whose proofs are too long and technical to be included here. Recall that a group A is called locally cyclic if every finite subset of A generates a cyclic subgroup of A. In particular, any locally cyclic group is Abelian. In view of a classical result due t o Ore (see e.g., Schmidt [95, Theorem 1.2.3]), a group A is locally cyclic if arid only if the lattice IL(,4) is distributive. Therefore, such groups are also called distributive groups, or shortly, D-groups.
a ( y ) and a ( x A y) = a ( x ) A a ( y ) , Vx, y E L.
It is easily seen that any lattice isomorphism between complete lattices commutes with arbitrary meets and joins. The morphisms in L a t will be denoted by small Greek letters a , B. 7 ,p, q!~,. . . . Clearly, any group morphisrn f : A + B yields a rnorphisrn in P o s which is not necessarily a morphism in L a t . A map a : L -+ L' is an isomorphism in L a t if and only if a is an order-preserving bijection such t h l ~ tits inverse a-' is also an orderpreserving map, in other words, a is an isomorphism in the category P o s . This irnplies that for any isomorphism f : A + B in G r , the map is an isomorphism in L a t . This fact can be described briefly by saying that we have a canoriical map u-here, if C is any category, then Isornc(X, Y) denotes the set, possibly empty, of all isomorphisms from the object X of C t o the object Y of C.
LF.\TM:I15.2.3. Let A be a distributive group and let B be a n y group. A lattice-isomorphism p : IL(,4) 4 IL(B) i s induced by a group isomorphism A % B if and only if 9 is index-preserving.
316
Chapter 15
PROOF. See Baer [27, Corollary 4.41 LEMMA 15.2.4. For any normal lattice-isomorphism A Zl, B , B is Abelian whenever A is Abelian. PROOF.See Baer [27, Theorem 7.11).
Recall that for any group A and for any prime number p we have denoted by tp(A) its pprimary component, that is. tp(A) = { a ~ A l o r d ( a )=p7' for some n~
N ) .
On the other hand, according t o Lemrna 15.2.5, B = cp(tp(A)), so, if we could show that for every p E P there exists a n isomorphism fp : tP(A) 2 cp(tp(A)), then clearly BPEP f P : A + B will provide a desired isomorphism. Thus, without loss of generality, we can suppose that A is a n Abelian p-group and B is an Abelian group which is lattice-isomorphic t o A. If A is a distributive group, then p = IL(f ) for some f E Isom Gr (A, B ) by Lemma 15.2.3, hence A Z B . If ,4 is not a distributive group, then apply 0 Lemma 15.2.6 t o obtain a n isomorphism A E B .
epEp
It is well-known that any torsion Abelian group A can be decomposed into the internal direct sum of its pprimary components: A = @t , ( ~ ) .
pEP
15.3. Infinite A b e l i a n G-Cogalois e x t e n s i o n s
See 1.1.2 for the concept of internal direct sum of a family of subgroups of a group. LEMMA 15.2.5. Let A be a torsion Abelian group, and let B be a group which is lattice isomorphic to A via cp : IL(A) -7 IL(B). Then B is a torsion group, and B =
In this section we investigate infinite Abelian G-Cogalois extensions. Mainly, we are interested t o find out how the Abelian groups K n e ( E / F ) and G a l ( E / F ) associated with a n arbitrary Abelian G-Cogalois extension EIF are related. In the case of finite extensions these two groups are isomorphic by Theorem 5.2.2, but not in a canonical way. Before stating and proving the rnain result of this section, we shall reformulate Theorern 5.2.2 and Theorem 6.1.7 together, dealing with finite Abelian G-Cogalois extensions, in a form which is suitable t o be generalized for infinite extensions.
eq,,
cp(to (A))
is an internal direct sum decon~positionof the group B into p-subgroups. PROOF. See Baer [27, 11.1, Theorem 2.41.
P~orosrrro 15.3.1. ~ Let E/F be afinite Abelian G-Cogalois extension with n = exp(G/F*). Then
Kne ( E I F ) r Ch (Gal ( E I F ) )
L ~ a n ~15.2.6. ,i Let A and B be two lattice-isomorphic Abelian groups. If A is an Abelian p-group which is not distrzbutive, then A E B . PROOF. See Baer [27, Theorem 11.8 (a3) and (c)].
gal ( E I F ) , p,,(E)).
PROOF. By Theorem 6.1.7, there exists a canonical group isomorphism Kne ( E I F ) 2 2' (Gal ( E I F ) ,p7,(E)), and by Theorem 5.2.2, there exists a non-canonical group isomorphisrn Kne ( E I F ) r Gal ( E I F ) . Further, the finite Abelian group Gal ( E I F ) is isomorphic, but not in a 0 canonical way, with its character group by Lemma 14.2.1 (4). REMARKS 15.3.2. (1) Observe that, by Lemrna 14.2.1 (2), Ch (Gal ( E / F ) ) = Hom(Ga1 ( E I F ) , p,,(C)) for m y finite Abelian G-Cogalois extension E/F with n = exp(G/F*). Since EIF is a n n-bounded Galois extension, we have gcd(n, e ( F ) ) = 1 and
THEOREM 15.2.7. Let A be a torsion Abelian group, and let B be a group which is lattice-isomorphic to A via cp : L(A) 2 L(B). If cp is index-preserving and normal, then A r B .
PROOF. First of all, note that B is Abelian by Lemma 15.2.4. Since A is a n Abelian torsion group, we can decompose A into the internal direct tp(A) of its pprimary components tp(A). The latticesum A = isomorphism cp : IL(A) & IL(B) obviously yields for every p E P a lattice isomorphism cp, : L(tP(A)) 7 IL(cp(tp(A))), which is also indexpreserving.
epEp
318
Chapter 15
<,, E E by Corollary 5.1.7, hence p,, ( E ) = p,,(Q) there exists a non-canonical group isomorphism
E prL (C). Consequently,
PROOF. Consider the following canonical maps yl

:
IL(Kne ( E / F ) )
--+ Subextensions ( E / F ) , H/F* I+ F ( H ) / F ,
Z1(Gal ( E lF ) , p,, ( E ) ) 2 Hom (Gal ( E I F ) , p,, ( E ) ) . Note that both Z1 (Gal ( E I F ) , p,,(E)) and Horn(Ga1 ( E I F ) , p,(E)) are subgroups of the group Maps(Gal( E I F ) , p,,(E)) of all maps from the group Gal ( E I F ) to the group p,,(E). Exercise 18 asks when these two isomorphic subgroups of Maps(Gal( E I F ) , p,, ( E ) ) coincide.
E F , that is, if E / F is a finite classical Kurnrner (2) If additionally i,, hence, every 1-cocycle f extension of exponent n, then p,,(E) = pTL(F), of Gal ( E I F ) with coefficients in p,,(E) = p,,(F) is actually a morphism of groups, since f ( u r ) = f (u) . u(f ( 7 ) ) = f ( u ) . f (T),for every a?r E G a l ( E / F ) . Thus,
79 : Subextensions ( E / F )
+ E ( G ~ ~ ( E / F ) )K , / F I+ Gal(E/K),
I+
y, : E ( G ~ ~ ( E / F + ) ) IL(Ga1( E / F ) ) , B
B'.
Since yl is an isomorphism of lattices by Theorem 11.2.6, yz is an antiisomorphism of lattices by the Fundamental Theorem of Infinite Galois Theory (Theorem 14.1.6), and ys is an anti-isomorphism of lattices by Proposition 14.1.5 and Lemma 14.2.5 ( I ) , we deduce that their composition 7=7 3 0~ '0 17 1 yields an isomorphism of lattices y : IL(Kne ( E / F ) ) and we are done.
-+ lL(Ga1 ( E / F ) ) , H J F * c, G ~ ~ ( E / F ( H ) ) ~ ,
C3
and then, we obtain a canonical group isomorphism Kne ( E /F )

E Horn
Kext, we are going to extend Proposition 15.3.1 t o infinite extensions. To do that, we need some preparatory results.
(Gal ( E / F ) , p,, (E)).
Note that this yields a group isomorphism Kne ( E I F ) E Cli (Gal ( E I F ) ) , which is not canonical since it depends on the chosen isomorphism of cyclic 0 p7,(C). groups p, ( E )
LEMMA 15.3.4. Let A E LCA and B E E(A). T h e n
o-+A/B+A^+E+o
is a n exact sequence i n the category LCA of all locally compact Abelian groups. I n particular, there exist canonical isomorphisms i n LCA
A
<
For the rest of this section we shall preserve the notation and terminology from Chapter 14. In particular, LCA denotes the category of all locally compact Abeliari groups. For any group A, the lattice Subgroups (A) of all subgroups of ,4 will be denoted shortly, as in the previous section, by L(A). Also, if A is a topological group, the lattice Closed Subgroups ( A ) of all closed subgroups of A4will be denoted by z(.4). As in Section 14.2, for any A t LCA arid ikny subsets 0 # X C A and 0 # Y & A^ we shall use the following notation:
A/B
E BL
and
,?/B~.
PROOF. See Bourbaki [39, T h h k r n e 2, Chap. 2 1 or Hewitt and Ross
[72, Theorem 23.25, Theorem 24.11, and Corollary 24.121.

LEMMA 15.3.5. Let E/F be a G-Cogalois extension. T h e n
I &/HI I = iF(H2) : F ( H l j ]
for any sub&oups H I , HP of G, with F* HI
<
< H2.
xi= { x ~ . ? ~ X ( a ) = V~ a ,t X } ,
~ ' = { a ~ ~ ~ ( a V ) X =E lY , }.
PROPO~IT 15.3.3. I O N If E/F is any Abelian G-Cogalois extension, then, the groups Kne(E/F) and Cli(Gal(E/F)) are lattice-isomorphic via the canonical lattice isomorphism
PROOF. The extension F ( H 2 ) / F is H2-Kneser by Proposition 11.1.2. Then, the extension F ( H 2 ) / F ( H 1 ) is F ( H 1 ) * H2-Kneser by Proposition 11.2.2, and F ( H l ) *n H2 = F ( H l ) n H2 = HI by Lemma 11.2.1. Again by Proposition 11.1.2, every set of representatives of
is a vector space basis, possibly infinite, of F ( H 2 ) over F ( H 1 ) . This implies the desired equality. 0
320
Chapter 15
15.3 Infinite Abeliar~G-Cogalois e x t e ~ ~ s i o r ~ s
32 1
LEMMA 15.3.6. Let K1 C K 2 be two intermediate fields of an Abelian extension E / F . If K2/K1 is a finite extension, then (Gal ( E / K 2 ) l : Gal ( ~ 1 ~ 1 ) '= ) [ K 2 : K1 1,
where
"' " is taken in the group Ch (Gal ( E I F ) ) .

0
PROOF. Let C1 C2 be two subgroups of K n e ( E / F ) = G / F * with the index (C2 : C1) finite. Then, there exists H I , H2 E IL(G/F*) such that F* 6 H1 6 H2 < G and Ci = H i / F * , i = 1, 2. We have to show that
( G : C1) = (y(C72) : y(C1)). By Lemma 15.3.5 one has (C2 : C 1 ) = ( H 2 : H I ) = [ F ( H 2 ) :F ( H I ) ] , so, it remains only to prove that (y(C2) : y(C1)) = [ F ( H z ) : F ( H l ) ] . Set Ki = F ( H i ) , i = 1. 2. Then y(Ci) = Gal(E/Ki))', hence (7(C2): r(C1)) = ( G ~ ~ ( E / K ~: )G ' ~(E/K~)')
= [ K 2 :K 1 ] = [ F ( H 2 ): F ( H l ) ]
PROOF. For simplicity, set A = Gal(E/F) and Bi = Gal(E/Ki), i = 1,2. By Lemma 15.3.4, the exact sequence of compact Abelian groups
--+ B 1 / B 2 + "I/B2 + -4/B1 +O
yields the exact sequence of discrete Abelian groups
Also, we have A/Bi "= Bf , i = 1,2. Since the diagram below
is clearly commutative, we deduce that
B$/B~ E B1/B2.
Since E/F is an Abelian extension, we have B1/B2 = Gal(E/K1)/Gal(E/K2)r Gal(K2/K1), by Theorem 14.1.6 (3). But K 2 / K 1 is a finite Abelian extension, hence its Galois group is a finite Abelian group, and so, by Lemma 14.2.1 (4), we have
by Lemma 15.3.6, and we are done.
THEOREM 15.3.8. If EIF is an Abelian G-Cogalois extension, then the discrete torsion Abelian groups Kne ( E I F ) and Ch(Ga1 ( E I F ) ) are isomorphic.
PROOF. By Proposition 15.3.3, the Abelian groups Kne ( E I F ) and Ch(Gal(E/F)) are lattice-isomorphic via the isomorphism of lattices
y : IL(Kne ( E / F ) ) defined by y(H/F*) = G ~ ( E / F ( H ) ) ~ , for every H / F * 6 Kne(E/F) = G / F * , where F* 6 H G. Since Kne(E/F) 6 Cog(E/F), and Cog(E/F) is precisely the torsion subgroup of the Abelian group E*I F * , it follows that Kne(E/F) is a torsion Abelian group. Further, the lattice isomorphism y is index-preserving by Lernma 15.3.7, and it is obviously normal, since Ch(Gal(E/F)) is an Abelian group. Now, apply Theorern 15.2.7 t o obtain the desired result.
--+
IL(Ga1( E / F ) )
LEMMA 15.3.7. If EIF is an Abelian G-Cogalois extension, then the isomorphism of lattices y : L(Kne(E/F)) --+ L(Ga1 ( E / F ) ) , ( H / F * ) ct G ~ ~ ( E / F ( H ) ) '
given b y Proposition 15.3.3 is strictly index-preserving.
RXMARKS15.3.9. (1) When applied to finite Abelian G-Cogalois extensions, the proof of Theorem 15.3.8 provides an alternate proof of Theorem 5.2.2.
322
Chapter 15
(2) A shorter approach t o the result of Theorem 15.3.8, which avoids the technical facts involving lattice-isomorphic groups and which is based on Theorem 5.2.2, could be the one presented below. Let EIF be an Abelian G-Cogalois extension. It is convenient t o index in an obvious way the directed set { H ( F * < H < G , H I F * isfinite) in order to obtain a direct system (Gi, pij)i,jEI of groups. Thus, for every i, j E I we set i j if and only if Gi E Gj, and for any i j we define the morphisrn pij : Gi + G j as being the canonical inclusion. Clearly
COROI,I,ARY 15.3.10. For any Abelian G-Cogalois extension E I F , the totally disconnected compact Abelian grou.ps Gal(E/F) and Ch(Kne(E/F)) are topologically isomorphic. PROOF. By Theorem 15.3.8, there exists a group isomorphism
f : Kne ( E I F ) 4 Ch (Gal ( E I F ) ) ,
which induces a topological isomorphism
f^: Ch (Kne ( E I F ) ) 2 Ch (Ch (Gal ( E I F ) ) )

By the Pontryagin Duality (see Theorem 14.2.2 ( I ) ) , the compact groups Gal ( E I F ) and Ch (Ch (Gal ( E I F ) ) ) are canonically isomorphic. 0 COROIJARY 15.3.11. If EIF is an Abelian Cogalois extension, then the discrete torsion Abelian groups Cog ( E I F ) and Ch(Ga1 ( E I F ) ) are isomorphic. PROOF. Any Cogalois extension EIF is T(E/F)-Cogalois by Theorem 12.2.3. Since K n e ( E / F ) = T ( E / F ) / F * = Cog(E/F), we can apply 0 Theorem 15.3.8 to obtain the desired isomorphism. REMARK 15.3.12. Another approach t o the proof of Theorern 15.3.8, which is sketched below, is due t o the anonymous referee of this book.
<
<
Kne ( E I F ) = G I F * = Further, for every i E I we have
U (Gi/F*) = lim (Gi/F*).

i61
iG
= Kne ( F ( G i ) / F ) 2 Gal ( F ( G i ) / F ) by Theorem 5.2.2. Kote that the two isomorphisms above are not canonical. Thus, we obtain the following isomorphisms of topological groups:
Ch (Kne ( E I F ) ) = Ch ( G I F * ) = Hom,(G/F*, T)
LEMMA A. Let E/F be an Abelian G-Cogalois extension. Assume that G I F * = $iEI(Gi/F*), and let Ei = F ( G i ) for every i E I . If Kne(Ei/F) g C h ( G d ( E i / F ) ) for all i E I , then K n e ( E / F ) EZ Ch(Gal(E/F)).
r lim Gal ( F ( G i ) / F ) r Gal ( E I F ) .
i';i
Kote that the last isotnorphistn follows from Proposition 14.1.5 (see also Exercise 3, Chapter 14) since EIF is an Abelian extension, hence, for every i E I there exists a canonical group isomorphism Gal ( F ( G i ) / F ) 2 Gal (E/F)/Gal (E/F(G;)). The approach above contains a serious gap which could affect its validity; namely, the two isomorphisms lim HomC(Gi/F*, T)
C-
n,,, Gal(Ei/F) with the product topology. Kote that each extension E i / F , i I , is an Abelian Gi-Cogalois extension. Since each continuous
E
PROOF. We have K n e ( E / F ) =
Kne(Ei/F) and Gal(E/F)
character of the product group kills almost all factors, we deduce that
Recall that for an Abelian group A and a prirne number p we denote by t,(A) the pprimary component of A. LEMMA B. Let EIF be an Abelian G-Cogalois extension with Kneser group G I F * a p-group, where p zs a prznle number. Assume that t,(p(F)) is a finite group. Then, the following statements hold. (1) If t p ( p ( E ) ) is a finite group, then G I F * is a group of bounded order.
iEI
E li~n Gal ( F ( G i ) / F ) E Gal CiE I
(EIF)
considered are not canonical, hence it is not clear that we really obtain a projective system of groups which is coherent in such a way as t o produce the Galois group Gal(E1F). 0
324
Chapter 15
1 5 . 4 lfxercises to Chapter 15
325
(2) If tP(p(E)) is an infinite group, then G I F * " ! t,(Q/Z) B is a group of bounded order.
@ B,
where
Ch(Gal(E/F)) = ZA(Gal(E/F),t,(p(E))), and it suffices t o apply Theo0 rem 15.1.4.
PROOF. (1) Let x E G and k = ord(2). Since F ( x ) / F is F* (2)Kneser, it follows that Irr(x, F ) = X k - xk, and so, all the conjugates of x over F are Cx, C running over p k ( F ) , and they rnust lie in E . Therefore, k Itp(p(E))I, which shows that G / F * is a group of bounded order.
<
(2) We denote by pPn; the pprimary cornponent t,(p(F)) of the group p ( F ) of all roots of unity in a,n algebraic closure F of F . Since t,(p(E)) is assumed to be an infinite group, then, by using Theorem 12.1.8, one shows that the group G / F * contains a subgroup Q isomorphic to the quotient group A d := /L,= /t,(p(F)). KOW,observe that A4 is isomorphic to the quasi-cyclic group Z,- := t,(Q/Z) of type p W . Since M is a divisible -4belian group, we can write G / F a = Q a ( H I F * ) for some F* H G. It remains to show tha.t B := H / F * is a group of bounded order. But K = F ( H ) cannot contain any ppower root of unity not already contained in F . Therefore, by the proof of ( I ) , exp(B) 6 Itp(p(F))I, and we are done.
1. Let EIF be an arbitrary Galois extension, and let K be any intermediate field of E / F . Show that the diagram below, where all the arrows are canonical morphisms,
< <
is commutative. 2. Let E/F be an arbitrary Galois Cogalois extension with Galois group l?. Prove that the canonical morphism is surjective for every closed subgroup 4 of T. (Hint: Use Exercise 1 and Theorem 12.2.4 (4).)
3. Give an example of infinite G-radical extension E/F for which p G ( E ) is a strict subset of p ( E ) .
4. Let EIF be a Galois extension, and let L I F be an extension such that E and L are subfields of some other field, E n L = F , and p ( L E ) = p(E). Let G be a group such that F* 6 G T(E/F), and denote GI = GL*. Prove the following statements. (a) G I F * G 1 / L a , and So, OGlp* = OGlll,*. (b) The extension F ( G ) / F is G-Kneser (resp. G-Cogalois) if and only if the extension L(Gl ) / L is GI-Kneser (resp. G I Cogalois). (Hint: Apply the Infinite Kneser Criterion, Theorems 12.1.12, and Theorem 15.1.9.)
PROOF O F THEOREM 19.3.8. By decomposing the Kneser group of the given Abe1ia.n G-Cogalois extension EIF into its pprirnary components, Lemma A shows that. there is no loss of generality to assume that E/F is an Abelian G-Cogalois extension with Kneser group G / F * a pgroup. There are three cues:
Case 1: If tP(p(E)) is a finite group, then G I F * is a group of bounded order by Lemma B and therefore it is a direct sum of a family (Gi/F*)iEl of finite cyclic groups by Theorem 1.4.2. Hence E is the compositum of the family of fields (E(Gi))iEl. By Proposition 15.3.1, Kne(Ei/F) Z C h ( G d ( E i / F ) ) for all i E I. Lemma A now yields the result. Case 2: If t p ( p ( F ) ) is a finite group but t,(p(E)) is an infinite group, , := F ( p p m ) then Lemma B allows us to write E as the compositum of F and K := F ( H ) , where F* 6 H 6 G and H / F * is a group of bounded order. By referring to the first case and using Lemma ,4 we reduce the In this case, K n e ( E / F ) E ZPm and G a l ( E / F ) is problem to E = F,. isomorphic to the additive group of the ring 2, of p a d i c integers. Note that-on needs to use that C2, E F for both of these statements. Since Ch(Z,) E Z,- by Exercise 13, Chapter 14, we are done. Case 3: If t , ( j ~ ( F ) ) is an infinite group, then E/F is a classical Kummer extension, and G a l ( E / F ) acts trivially on t,(p(E)). Consequently,
<
5. Let E/F be a Galois extension, and let L / F be an extension such that E and L are subfielcls of some other field and E n L = F . Let K be a subfield of E containing F such that K L / L is a GI-radical extension and C,, E E for all nl E Oc:,/,,*, and set G = G I n E * . Prove the following assertions.
Chapter 15
(a) The extension K l F is G-radical and G I F * S G I IL*. (b) The extension K I F is G-Kneser (resp. G-Cogalois) if and only if the extension K L I L is GI -Kneser (resp. G I-Cogdois). (Hint: Apply Corollary 15.1.3, Theorem 15.1.9, and Exercise 4.)
is i~ljective for all A, B t C. 15. We say that the groups A and B are lattice-anti-isomorphic if there exists a lattice anti-isomorphism between IL(,4) and IL(B). Investigate the properties of lattice-anti-isomorphic groups. 16. (Baer-Sadovskii Theorem). Let A and B be two lattice-isomorphic groups, and let y : IL(,4) -+ IL(B) be an isomorphism of lattices. Assume that there exists a nonernpty set U of subgroups of A satisfying the following three conditions. (a) Every H E U is finitely generated. jb) For every a E A there exists H E U such that a E H . (c) For any H I , H2 t U there exists H3 E U with H I U Hz & H3. If for every H E U , yII1 is induced by an isomorphism of H onto i p ( H ) ,then prove that q is induced by an isomorphism f : A + B . (Hint: See Schmidt [95, Corollary 1.3.71.) 17. Let EIF be an Abelian G-Cogalois extension. By Theorem 15.3.8, there exists a non-cmonical group isomorphism
6. Let EIF be a Galois extension, and let L I F be an extension such th& E and L are subfields of some other field and E n L = F . Let K be a subfield of E containing F such that K L I L is a G I radical extension with G I / L * a group of exponent n, and assume that Cn E E . If we set G = G I n E*, then prove the following assertions. (a) The extension K / F is G-radical and G I F* G I I L*. (b) The extension K I F is G-Kneser (resp. G-Cogalois) if and only if the extension K L I L is G1-Kneser (resp. GI -Cogalois) . (Hznt: Observe that OG', / , , . = IDnr and use Exercise 5.)
"
7. Assume that all fields below are subfields of some other field. Prove the following statements. (a) If El / F arid E2I F are Abelian extensions, then so is El E 2 / F . (b) If EIF is an Abelian extension and K I F is any extension, then E K I K is an Abelian extension. (c) Any subextension and any quotient extension of an Abelian extension are Abeliari extensions.
8. Prove that a group G is finite if and only if the lattice IL(G) is finite.
E Isom Lat(IL(,4),IL(B)), where A and B are lattice9. Let y , isomorphic groups. Show that y = $ if and only if y ( X ) = @ ( X ) for every cyclic subgroup X of A.
f : Kne ( E I F ) 7 Ch (Gal ( E I F ) ) ,
and by Proposition 15.3.3, there exists a canonical lattice isomorphism
y : IL(Kne ( E I F ) ) -7 IL(Ch (Gal ( E I F ) ) ) .
When is 7 = lL( f ) ? 18. Characterize those n-bounded Abelian G-Cogalois extensions EIF for which the isomorphic groups Z1(Gal( E / F ) , p , , ( E ) ) and Hom(Ga1 ( E lF ) , p,, ( E ) ) are equal.
10. Prove that a group is distributive if arid only if it is isomorphic t o a subgroup of Q or of Q/Z. 11. Prove that a finite group is distributive if and only if it is cyclic.

Section 15.1. Except for Theorem 15.1.2, which is due t o Barrera-Mora, Rzedowski-Calder6n and Villa-Salvador [30], all of the results of this sec-
12. (Baer [27]). Let A be a group. Prove that A
El, Z
w A r Z.
13. (Baer [27]). Let r, n l , . . . ,n, E N*, and let p l , . . . ,pT E I P be distinct primes. If A is a cyclic group of order p;" . . . . p : ? and B is any group, then prove that A SL B if and only if B is a cyclic group of order 97' - . . . . qF7. with distinct primes ql, . . . ,q, E I F ' . 14. Find classes C of groups for which the map LA. I j : Isom G , (A, B )
tion are due to Albu [9].

Section 15.2. This section contairls the basic terrnirlology and notation for lattice-isomorphic groups, as well as: the presentation without proofs, of some technical results on this material. We have mainly followed Baer [27] and Schmidt [95]. Theorem 12.1.7 is only mentioned en passant in the introduction of the fitnr1;unental paper of Baer [27], but it is neither
---+ Isom Lat (IL(,4),IL(B)), f
* IL(f ),
328
Chapter 15
explicitly stated nor proved in the rest of that paper. Surprisingly, we could not find it in other sources, though it provides an important case when lattice-isomorphic groups are isomorphic. According to a result established by Whit~narl in 1946 (see e.g., Griitxer [62, Corollary 5, Chap. 4, 4]), every lattice is isomorphic t o a sublattice of IL(G) for some group G. The term lattice-isomorphism has various other names in the literature: subgroup-isomorphism in Baer [27], projectivity in Schmidt [95] and Suzuki [ l o l j , L-isomorphism, etc. The distributive groups are called ideal-cyclic groups in Baer [27, p. 81, D-groups in Suxuki [101], and locally cyclic groups in Kurosh [78].
Section 15.3. The results of this section are due to Albu [lo], and Albu and Basarab [15]. Remark 15.3.12 reproduces the completely different approach to the proof of Theorem 15.3.8 which is due to the referee of this book and to whom the author is very thankful. The version in the framework of abstract Cogalois Theory of the group isomorphism given by Theorem 15.3.8 is discussed in ,41bu and Basarab j17j.
Bibliography
M. A C O S T A Dl3 O R O Z C O and W.Y. V ~ L E Z The , lattice of subfields of u radical extension, J. Number Theory 15 (1982),388-405. M. A C O S T A Dl? O R O Z C O and W.Y. V ~ T , R Z The , torsion group of a field defined by mdicals, J. Nurr~berTheory 19 ( l 9 8 4 ) , 283-294. 1 T . A L B L , Kurnmer extensions with few ruots of unity, J . S u m b e r Theory 4 (1 992), 322-358. T . AT,BL, From the irrationality of sums of radicals to Kneser and G-Cogalois Gaxeta Materrlatic5 100 ( 1995), 42 1-430. field extensions ( I ) (in Rornartiar~), T . A L B U , From the irmtionulity of sums of radicals to Kneser and G-Cogalois , Ganeta Matematic5 100 (1995), 61 1-619. field extensions (11) (in Ronrar~iar~) T. A L B U ! Field extensions having a Coqalois correspondence - A survey, in "Proceedirtgs o f the A rtrtual Meeting o f t he Romaniart Society o f Mathematical Sciences, Bucharest, 1 9 9 7 , Torrte I , Fds. XT. Tlechear~u, 1. D. Ion, and A.Verr~escu,1998, pp. 13-22. T . 4TJBL> Some exarnples i n Cogalois T h w r y with applications to elementary Field Arithmetic, J . Algebra Appl. 1 (2002), 1-29. T . AT,BC, Infinite field extensions with Co!lalois tmrrcspondent:t!:Corr~rr~. Algebra 30 (2002), 2335-2353. T . ACBU, Infinite field extensions with Galois-Cogalois w r r w p o n d e i w (I)! Rev. R o u n ~ a i n eMath. Pures Appl. 47 (2002),t o appear. T . A L B U , Infinite field extensions with Galois-Cogalois correspondence (IT), R e v . Rourr~air~e Math. Pures Appl. 47 (2002),t o appear. T . AJ,T)U, Field extensions with the unique subfield property, and G-Cogulois extensions, Turkish J . Math. 27 (2003),t c appear. T . AT213C, Corrigendum and Addendum to m y paper t~ncernin!qKurnmer extensions with few roots of unity, J . Surr~berTheory, t o appear. T . A L B U , O n radical field extensions of prime exponent, J . Algebra Appl. 2 (2003), t o appear. Clifford extensions, and Ilopf alqebras, subT . A L B U , Infinite Cogalois T l i t ~ r y , mitted. !~r.oups,and infinite Abelaan G T . .4T,BU artd 3. BAS.4 R A B , Lattic:c:-isornor'yI~it: Co!lulois field eztensioiis, J . Algebra Appl. 1 (2002), t o appear. T . AT,BU and 3. B A S A l t A B , Towuid a n abstruct Cogalois Theory ( I ) , submitted. T . AT,TlC artd 3. T l A S A R A B , Toward an abstruct Cogalois Tl~wr:y(TI), in preparatior~.
T . AT,BU arld C. SASTASI?SCU, "Relutiue Finiteness in Module Theory", Marcel Dekker, Irrc., Sew York a r ~ d Basel, 1984. T . AT,BC a r ~ d F . STCOT,AI?, Kneserfield extensions with Cogulois corwspondence, J . Surr~ber Theory 52 (1995), 299-318. T . ALDU and F . ?JCOT,AI?, G-Cogalois field extensions and prirrlitive elements, irl "Syrnposia Gaussiana", Corlference A: Matherrlatics a r ~ dTheoretical Physics, Eds. M. Rehara, R. Fritsch, and R.G. Tintx, Walter de Gruyter & Co., Berlin Xew York, 1995, pp. 233-240. T . ALRU arrd F. SICOT,AI?, Ileckesche Systerne idealer Zahlen und Knesersche Kurpe7r:rweiterur~ger1, Acta Arith. 73 (l995), 13-50. T . ALRU and F . XICOLAE, Finite radical field extensions and crossed hornornorphisrns, J . Turr~berTheory 60 (1996), 291-309. T . AT,RT;, F. STCOT,AE, and M. T E S A , Some i ~ r n u r k son G-Cogalois field extensions, Rev. Rourr~airleMath. Pures Appl. 41 (I996), 145-153. T . ALBU arrd 1,. PASAITOPOT,, Quurtic field extensions with no proper interrntdiate field, Rev. Rourrrair~e Math. Pures Appl. 47 (2002), to appear. T . 41,BU and \,I. T F S A , Infinite Coqulois Theory, hlathernatical Reports 3 (53) (200 1 ), 105-132. l?. ARTIS, " Guloissche Theorif!", B.G. Teubrler Verlagsgesellschaft, Ixipzig, 1959. R. BAER, The si!/nificxnce of the system of subgroups for the structure of the !y.oup. Arner. J . Math. 61 ( l939), 1-14. A . BAKFR arrd TT.%l. START<, On u fundamental inequality in number theory, Atlrr. of Math. 94 jl971), 190-199. P. BAR.RI?RA-LlORA, 0 7 1 subfields of radical r:xtensions, Corr~rn.Algebra 27 (l999), 464 1-16/19. F. BARRERA-MORA, XI. RZRDOWSKI-C:AT,DER~S, arld G . VTT,T,ASAT,VADOR, On Cogalois extensions, J . Pure Appl. Algebra 76 ( L99l), 1-1 1. F . BARRERA-MORA, M. RZEDOWSKI-CAT,DRR~?;, and G. VIT,T,ASALVADOR, Allowable groups and Cogalois extensions, J . Pure Appl. Algebra 104 (l995), 1-11. F. BARRERA-MORA and W.Y. vI?T,EZ, Some results on rndical extensions, J . Algebra 162 (I 993), 295-30 1 . J.R. RASTIDA, "Field Extensions und Galois Tl~eory",Adclisorl-Wesley Publishing Company, Reading, Massachusetts, 1984. I?. BKCKER, R. GROBE, arrcl M. STI?RMASS, Radicals of binorniul ideuls, .J. Pure Appl. Algebra 117 & 118 (I997), 41-79. A. BESTCOVTTCTI, 0 7 1 the linear. independence of fractional powers of integers, .I. I,ondorl Math. Soc. 15 (1940), 3-6. G. BTRKTTOFF, "Lattice Theory", Arnericarr Mathematical Society Colloquium Publicatiorrs, Vol. X X V , 3rd Editiorl, Providerrce, 1984. Z.I. BORl?VTTCTT and I.R. STTAFAREVITCTT, "Nurnber Tlleory", Academic Press, S e w York, 1966. S . BOURBAKI, " Topoloqit: Gtntruk:", ITerrnarrrr, Paris, 1966. S . BOURBAT<I, " The'oiie Spt:cti.clles", TTerrnarlrl, Paris, 1967. S . BOCRBAKT, "Alqdbre"~Chapitres 1 a 7. Masson. Paris, 1981.
A. BRANDIS, ~ b e die r multiplikatiue Struktur uon Kiirpererweiterur~ger~, Math. Z. 87 (1965), 71-73. S. CAESEPEET,, "Rruuer Groups, IIopf Algebms, and Galois Theory", Kluver Academic Publishers, Dordrecht, 1998. A. CAPET,T,I, Sulla riduttibilitd delle equazioni ulgebriche I , Rerld. Accad. Sci. Fis. Mat. Napoli 3 (1897), 243-252. A. CAPET,LI, Sulla riduttibilitd delle quuzioni algebriche 11, Rend. Accad. Sci. Fis. Mat. Napoli 3 (1898), 84-90. A. CAPET,LI, Sulla riduttibilitd della funzione x7' - A in urc, camyo qualunque di razionalitd, Math. Ann. 54 (1901), 602-603. Number* Theory", Academic J.W.S. CASSETS and A. F R ~ I T L T C I I "Algebraic , Press, London S e w York, 1967. S.L. CIIASE arrd M. SWEEDT,ER, "llopf Algebrns and Gulois Theory", Lecture Sotes in Mathematics, Vol. 97, Springer-Verlag, Berlir~ TIeidelberg ?;ew York, 1969. 1,. CHTT,DS, "A Concrete Introduction to Iligher Algebra", Springer-Verlag, S e w York, Tnc., 1979. P. CRAW1,EY and R.P. DIT,WORTTT, "Algebraic Theory of Lattices", PrenticeTTall, Trlc., Erlglewood Cliffs, Yew .Jersey, 1973. E.C. DADE, Cornpoundiilg Clifford's Theory, Arlrl. of Math. 91 (1970), 236-290. E.C. DADE, lsornorphis7r~sof Clifford extensions, Arrrl. of Math. 92 (1970), 375433. E.C. DADE. Grwqqrnded rings and modules, Math. Z. 174 (l980), 24 1-262. G. DARDT, Sullu riducibilitd delle t!quuzioni al!/ebiiche, Ar~rl. Mat. Pura Appl. ( I ) , 4 (1926), 185-208. S. n A s c A r , n s c v , C. sAsTAsEsCU, and 5. RATASU, "llopf Algebras: An lntroduction", Marcel Dekker, Tnc., Xew York and Basel, 2001. D.S. DUMMTT, On the toi.sion in quotients of the rrlultiplicatiue groups 271 Abelian extensions, in: "Nurnber Theory", Proceedirrgs of the Tr~terrlatior~al Surr~ber Theory Corlference held at Universit6 T,aval, 1987, Eds. J-M. De Korrirrck arld C. Levesque, Walter de Gruyter, Berlirl arrd New York, 1989. E.E. ENOCTTS, J.R. GARCIA ROZAS, and 1,. OYOKARTE, Corrtyact coGalois groups, Math. Proc. Carrhridge Philos. Soc. 128 (2000), 233-244. I?..E. ENOCHS, J.R. GARCIA ROZAS, arld I,. OYOSARTE, Covering rr~orphisms, Cornrn. Algebra 28 (2000), 3823-3835. 1,. GAAT,, "C1assic;al Galois Theory with Examples". Markharn Publishing Co., Chicago, 1971. D.J.11. GARLING, "A Course in Gulois Theory", Carnbridge Ur~iversityPress, Cambridge, 1986. D. GAY and W.Y. vI?T,EZ, On the dqree of the splitting field of a n irreducible birwrniul, Pacific J . Math. 78 (1978), 1 17-120. D. GAY and W.Y. vI?T,EZ, The torsion group of a radical extension, Pacific J . Math. 92 (I981), 317-327. G. G R ~ ~ T Z E R "Geneid , Lattice Theory", Birkhluser Verlag, Basel and Stuttgart, 1978. C:. GREITTIER arrd D.K. TTARRTSOS, A Gulois t~rwspndenc:efor radical exterlsions of fields, J . Pure Appl. Algebra 43 (l986), 257-270.
Index
Abeliar~exteosion, 38 adjur~ctior~ field - , 24 ring - , 23 algebra, 242 corr~odule- , 247 fir~itelygenerated - , 23 graded - , 229 TTopf - , 247 rr~otlule- 247 - rnorphisrn, 243 - of finite type, 23 strongly graded - , 230 algebraic - closure, 27, 28 degree of an - number field, 208 - elernent, 26 - extension, 27 - integer, 209 - number, 26, 209 - r~urr~ber field, 22, 208 algebraically closed field, 28 alrrlost G-Cogalois extension, 108 anti-rnorphisrr~ - of lattices, 35 - of posets, 34 ar~tipode, 247 secor~d - , 297 characteristic, 20 - exponent, 20 Chinese Rerriair~derTheorem, 18 class Hilbert - field, 221 - r~urnber, 211, 212 classical - Kurnmer exter~sion,174 - n-Kurnrner extension, 174 Clifford exter~sior~, 230 Clifford-Cogalois extension, 239 9l closed elerner~t, closure algebraic - , 28 integral - , 208 coalgebra, 243 - rr~orphisrn, 244 cobour~dary,154 cocycle, 154 - , 307 cor~tir~uous Cogalois - connection, 90 - exter~sion, 54, 276 exter~sior~ with - correspor~dence,94 G- - extension, 104, 269 - group, 54 - object, 91 star~dard - conr~ection, 93 cohornology group, 154 co~~tir~uo u s, 307 coir~mriar~ts, 2/18 cornodule, 245 - algebra, 247
bialgebra, 247 - r ~ ~ o r p h i s r247 r~, bour~tled group of - order, 17 - exterlsiol~,126 character group, 296
Index
- rnorphisrr~, 246 complete lattice, 34 cornpositurn, 25 condition - C o ( y a ) , 44 - C I ( n ;a ) , 44 - C , (71;a1,. . . , a ? ) , 201 conjugate elements, 29 cormectior~ Cogalois - , 90 Galois --, 90 corlorrnal exterlsior~, 88 corltirluous - cocycle, 307 - cohornology group, 307 - crossed hornornorptlisrn, 307 - ITilbert's Theorern 90, 307 coseparable extensior~,88 Criterion Gay-VClez - , 73 Gelleral Purity - , 270 Greitller-ITarrisor~- , 72 Tr~fil~ite Gay-Velez - , 280 lrlfirlite Greither-IIarriso~l- 276 lrlfir~iteKrleser - , 261 Infirlite n-Purity - , 275 Krleser - , 60 - , 102, 275 TL-Purity Vahlerl-Capelli - , 43 crossed homomorphism, 154 corltir~uous- , 307 cubic - extensiorl, 22 - rlurnber field, 22 cyclic exter~siorl, 38 cyclotornic - exter~sior~, 38 - field, 38 - polyrlorrlial, 38
of inseparability, 31 separable - , 3 1 derivative, 30 direct irlterrlal - sum, 17, 229 - product of groups, 17 - surrl of groups, 17 distributive group, 315 Doi-ITopf module, 247 domain, 208 Dedekirltl - , 210 integrally closed - , 208 prir~cipalideal - , 210 unique factorizatior~- , 21 0 Duality Porltryagirl - , 298 - Theorern, 297
-
Dedekirld dornairl, 2 I 0 degree inseparable - , 32 - of an algebraic r~urrlberfield, 22, 208 - of an elernent, 26 - of an exterlsiorl, 2 ' 1
elernerlt algebraic - , 26 closed - , 91 - , 230 tlornoge~~eous integral - , 207 prime - , 210 primitive - , 24 purely inseparable - , 32 separable - , 3 l trarlscerlderltal - , 26 exponent characteristic - , 20 - of a group, 48 exterlsiorl Abeliar~- , 38 algebraic - , 27 alrr~ostG-Cogalois - , 108 bourlded - , 126 classical Kurnrner - , 174 classical n-Kurnrner - , 174 Clifford - , 230 Clifford-Cogalois - , 239 Cogalois - , 54, 276 cor~orrnal- , 88 coseparable - , 88 cubic - , 22 cyclic - , 38 - , 38 cyclotorr~ic field - , 21
firlite - , 22 firlite ring - , 208 finitely generated - , 24 G-Cogalois - , 104, 269 G-Kneser - , 57, 260 G-radical - , 54 Galois - , 33 r-Clifford - , 230 r-Clifford-Cogalois - , 239 generalized Kurnrner - , 178, 285 generalized n-Kurnmer - , 178, 285 1 1 - - , 250 11-Galois - , 250 infinite - , 22 infinite Cogalois - , 276 infinite G-Cogalois - , 269 infinite G-Kneser - , 260 illfinite generalized Kurnrrler - , 285 infinite generalized TL-Kurnrner- , 285 irlfirlite Krleser - , 260 itlfirlite n - K u n ~ r r ~ e r with few roots of unity, 286 infinite quasi-Kurnrner - , 287 ir~firlitestrorlgly G-Kneser - 265 irlfir~ite strorlgly Kueser - , 265 integral - , 208 Kneser - , 57, 260 Kurr~mer - , 174 -s, 25 lir~early dis,joir~t TL-bounded- , 126 n-G-Kr~eser - , 187 - with few roots of unity, TL-Kurnrner 180, 286 - , 101 TL-pure norrnal - , 29 - of firlite type, 24 pure - , 70 purely inseparable - , 32 Q-pure - , 270 quadratic - , 22 quartic - , 22 - , 18 1 , 287 quasi-Kurnrr~er quotierlt - , 21 radical - , 54 repeated radical - , 170
ring - , 208 separable - , :3 I simple - , 24 sirnple radical - , 54 strictly alrnost G-Cogalois - , 108 strongly G-Krleser - , 97, 265 stror~gly Kneser - , 97, 265 trarlscenderltal - , 27 -with G/F'-Cogalois correspondence, 94 - with r-Galois correspondeoce, 94 - with the unique subfield property, I2 1 F-autornorphisrn, 21 P-isornorphisrn, 2 1 F-rnorphisrn, 2 1 field, 19 - adjunction, 24 algebraic rlurr~ber - , 22, 208 algebraically closed - , 28 cubic nurnber - , 22 - , 38 cyclotorr~ic - exter~sion. 21 firlite - , 28 fixed - 34 TIecke - , 219 IIilbert class - , 221 - , 21 ir~terrnediate rlurnber - , 22, 208 - of ir~variar~ts, 34 perfect - , 20 prirne - , 20 quadratic rlurnber - , 22, 209 - , 22, 135 quartic r~urr~ber splitting - , 27 - with few 71-th roots of unity, 71 - with few roots of unity, 71 firlite - extensiorl, 22 - field, 28 - ring exterlsion, 208 - topology, 293 finitely generated - algebra, 23 - exterlsiorl. 21 fixed field, 34
Index
Index
- n-Kurnrner extension, 178, 285 - neat presentation, 190
fractior~al ideal. 209 irwertible - , 209 - of a number field, 212 principal - , 209 Frobenius rnorphisrn, 20 Fundarnental Theorern of Finite Galois Theory, 35 Fundamer~talTheorem of Infinite Galois Theory, 295
- class group, 2 1 1, 21 2
- of posets, 35
Kneser - Criteriorr, 60 - extensiorl, 57, 260 - group, 107, 275 Ir~fir~ite - Criterion, 261 strongly - extension, 97, 265 strongly G- - extension, 97, 265 Krull topology, 293 Kurnrr~er classical - extension, 174 classical n- - extension, 174 - extension, 174 - exter~sior~ with few roots of unity, 180, 286 generalized - extension, 178, 285 178, 285 generalized TL- - extensior~, TL- - extensior~ with few roots of unity, 180, 286 quasi - extension, 181, 287 - Theory, 173, 283 iattice, 34 - anti-rnorpt~i~rr~, 35 corr~plete - , 34 - rnorphisrr~, 35 lattice-isornorpt~ic groups, 3 15 lattice-isorr~orptrisrr~, 3 15 index-preservir~g- , 3 15 oorrnal - , 31 5 strictly index-preserving - , 315 linearly disjoint extensions, 25 locally cyclic group, 315 rnaxirnal ideal, 2 10 rr~irrirnalpolynomial, 26 rnodule, 244 - algebra, 247 Doi-Tlopf - , 24 7 graded - , 230 (11, A)-TIopf - , 247 - rnorphisrn, 215 strorrgly graded - , 230 rnorph isrr~ F- - , 21 Frobenius - , 20
Galmost - Cogalois extension, 108 - Cogalois exter~sior~, 104, 269 - Kr~eserextension, 57, 260 - radical extension, 54 strictly almost - Cogalois exter~sior~, 108 strongly - Kr~eser exter~siorr, 97, 265 G-Cogalois a l ~ ~ ~o sexter~sior~, t 108 - exter~siorr,104 strictly alrr~ost- exterlsior~,108 G-Kr~eserextensior~,57, 260 G-radical exter~sior~, 54 T'-Clifford exter~sion,230 , 239 r-Clifford-Cogalois exter~sior~ Galois - conuection, 90 - extension, 33 extension with - corresporderrce, 94 - group, 21 11- - extension, 250 - 11-object, 247 - object, 91 92 standard - cor~r~ectiorr, Galois Theory Fundarnental Theorern of Finite - , 35 Furrdarnerrtal the or en^ of Infirrite - , 295 Infinite - , 295 Gav-Vklex - Criterior~, 73 - Criterion, 280 Tr~fir~ite General Purity Criterion, 270 ger~eralized - Kurr~rr~er exter~sior~, 178, 285
graded - algebra, 229 - rnodule, 230 - ring, 230 Greither-Tlarrisor~ - Criterion, 72 Infinite - Criterion, 276 group ct~aracter- , 296 Cogalois - , 54 cohornology - , 154, 307 distributive - , 315 exponent of a - , 48 Galois - , 21 ideal class - , 21 1, 212 - , 107, 275 Kr~eser locally cyclic - , 315 n-bounded - , 48 - of bounded order, 17 - of characters, 296 profinite - , 291 torsiorr - 47
- class of a rrurr~berfield. 212 - , 209 fractior~al

- of a r~urnberfield, 212 fractior~al integral - , 209 rr~aximal- , 210 prime - , 210 principal - domain, 210 principal fractior~al - , 209 independent family of subgroups, 17 index-preserving lattice-isornorphisrr~, 3 15 infinite - Cogalois extension, 276 - extension, 22 - G-Cogalois extension, 269 - G-Kneser extension, 259 - generalized K u n ~ r r ~exter~sior~, er 285 - generalized TL-Kumrner exterrsior~, 285 - Kneser exter~sior~. 259 - n-Kutnrr~er exter~sior~ with few roots of unity, 286 - quasi-Kurnrner exterrsior~, 287 - stror~gly G-Kneser exter~sior~, 265 - strongly Kneser exter~sior~, 265 lr~firriteGalois Theory, 295 Infinite Gay-V6lez Criterion, 280 Infinite Greither-TTarrisor~ Criterion, 276 Infinite IIilbert's Theorem 90, 307 Infinite Kr~eserCriterior~, 261 Infinite TL-Purity Criterior~, 275 inseparable degree, 32 integral - closure, 208 - element, 207 - extension, 208 - ideal, 209 ir~tegrallyclosed dornain, 208 ir~terrnediatefield, 21 irrterr~aldirect surn, 17, 229 ir~var iarrts field of - , 34 11- - , 251 ir~vertible fractional ideal, 209 iso~norphis~r~ F- - , 21
II- extension, 250 - Galois extension, 250 Galois - object, 247 - ir~variants, 251
Hecke - field, 219 - system of ideal nurnbers, 219 TTilbert class field, 221 TTilbert's Theorern 90, 154 cor~tirruous - , 307 irrfir~ite- , 307 hornogerreous - cornpor~ent,230 - element, 230 IIopf - algebra, 247 (IT, A)- - rr~odule,247 - rnodule, 247 - rnorphisrn, 247 ideal
algebras, 243 bialgebras, 247 coalgebras, 244 colnodules, 246 1Topf algebras, 247 lattices, 35 - of modules, 245 - of posets, 34 rrlultiple root, 30
7 L -
- of - of - of - of - of - of
- bounded - extension, 126 - group, 48 - bour~dedextension, 126 - bounded group, 1 8 174 classical - Kurnrner exter~sior~,
- G-Kneser extension, 187

generalized - Kurrlrrler exter~siorl,178, 285 - Kurrlrr~erexter~siorlwith few roots of unity, 180, 286 - pure exterlsior~,101 - Purity Criterion, I02 neat preserltatiorl, 190 generalized - , I90 Soetheriarl ring, 2 10 rlorln, 33 normal - closure, 29 - extension, 29 r~orrnallattice-isorr~orphisrn, 3 15 rlurr~ber algebraic - , 26, 209 class - , 21 1 - field, 22, 208 cubic - , 22 quadratic - , 22, 209 quartic - , 22 object Cogalois - , 91 Galois - , 91 Galois 11- -, 247 opposite poset, 34 owrfield, 21 overrirlg, 18
perfect field, 20 PTD, 210 polyrlorr~ial cyclotorrlic - , 38 rninirr~al - , 26 separable - , 30 Porltryagirl Duality, 298 poset, 34 - anti-isornorphisrr~,35 - anti-rnorphisrn, 34 - isornorphisrn, 35 - rnorphisrn, 34 opposite - , 34 prime - element, 210 - field, 20 - ideal, 210 - subfield, 20 prirnitive - elerrierlt, 24 - 71-th root of unity, 37, 211 prirlcipal - fractiorlal ideal, 209 - ideal dornair~, 210 profinite group, 29 1 pure extension, 70 purely ir~separable - elerr~ent, 32 - extension. 32 &-pure exterlsiorl, 270 quadratic - extension, 22 - rlurnber field, 22, 209 quartic - extensioo, 22 - rlurrlber field, 22, 135 quasi-Kurnrner exterlsion, 181, 287 quotient extension, 21 radical - exterlsion, 54 G- - extension, 54 71-th - , 54 repeated - exterlsion, 170 repeated radical extension, 170 ring-
- adjur~ction, 23 - exter~siorl.208 graded - 230 - , 210 Soetheriar~ - of algebraic integers, 209 strongly graded - , 230
root multiple - , 30 - of unity, 37, 211 primitive 71-th - of unity, 37, 21 1 simple - , 30 separable - degree, 31 - elerrlent, 31 - extension, 31 - polynomial, 30 set of representatives, 18 sigma r~otatioo,243, 246 simple - exterlsion, 24 - radical exteosion, 54 - root, 30 splittir~g field, 27 square-free, 209 star~dard - Cogalois conr~ectior~, 93 - Galois corlrlection, 92 Steir~itz'sExtension Theorern, 28 strictly alrnost G-Cogalois exter~sior~, 108 strictly index-preserving lattice-isorr~orphisrr~, 315 strong1y - G-Kneser extension, 97, 265 - graded algebra, 230 - graded module, 230 - graded ring, 230 - Kneser extension, 97, 265 subexter~sion,21 Theorem Chinese Rerrrair~der - , 18 Duality - , 297 Fur~darr~ental - of Fir~ite Galois Theory, 35 Fur~da~r~erltal -of lt~firlite Galois Theory, 295
ITilbert's - 90, 154, 307 Steinitx's Fxterlsion - , 28 topology finite - , 293 Krull - , 293 torsion group, 47 Tower Taw, 23 tower of fields, 23 trace, 33 transcerlder~tal - element, 26 - exter~sion, 27 UPD, 210 unique factorizatiorl dorr~air~, 210 unique subfield property, 121 U P , 121 Vahlen-Capelli Criterion, 43

Cogalois Theory

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cogalois Theory

Uploaded by

Copyright:

Available Formats

COGALOIS THEORY

PURE AND APPLIED MATHEMATICS

MONOGRAPHS AND TEXTBOOKS IN PURE AND APPLIED MATHEMATICS

Zuhair Nashed University of Delaware Newark, Delaware

56. 57. 58. 59. 60. 61. 62.

MARCEL DEKKER, INC.

To my beloved wife Marina

Copyright O 2003 by Marcel Dekker, Inc. All Rights Reserved.

PRINTED IN THE UNITED STATES OF AMERICA

FINITE COGALOIS THEORY

INFINITE COGALOIS THEORY

Chapter 15. 15.1. 15.2. 15.3. 15.4. 15.5.

"fi+...+ " G E Q M "&EQ Besides the extension Q ("C/aT, . . . , "*)/Q

forall l < i < r .

FINITE COGALOIS THEORY

= the set of all complex numbers

= the greatest common divisor of m and n

lcm(m, n) = the least common multiple of m and n q ( n ) = the Euler function of n

1.1 General notation a n d terminology

P n ID7, for any n E N

1 1Gi I xi = ei for all but finitely many i E I }

the (external) direct sum of an arbitrary family

= the internal direct sum of an independent

ST" Subgroups (G) (M)

1.2 A short review of basic field theory

E R ) = the left ideal of R

1.1.4. VECTOR SPACES

1.2. A short review of basic Field Theory

1 . 2 A short review of basic field theory

and only if its Frobenius morphism p~ is actually an isomorphism.

If Char(F) > 0, then it is necessarily a prime number.

A field F is called perfect if F = ~ " 0 i.e., , if every element of F is an e(F)-th power in F .

if P(F) G E. if P ( F ) G 0. Frobenius morphism. For any field F of characteristic p Char ( F ) =

> 0, the map

set of all subextensions of E / F . Note that Subextensions ( E / F ) is a poset,

1.2 A short review of basic field theory

1.2 A short review of basic field theory

Now, if A is an arbitrary subset of El then

where FAdenotes the set of all finite subsets of A.

1.2 A short review of basic field theory

Algebraic extension. Let E/F be an extension. We say that E/F is an

the evaluation map at u defined as

4 f ) =f is a ring morphism, with image

1.2.5. SPLITTING FIELD

1 . 2 A short, review of basic field theory

1.2 A short review of basic field theory

A polynomial f E F [ X ] is said to be separable if all its irreducible factors in F [ X ] are separable.

1.2.8. SEPARABLE EXTENSION

(k times). Thus, one obtains a map,

D : F [ X ] +F [ X ] , f which has the following properties.

1.2 A short review of basic field theory

we define the following elements:

1.2 A short review of basic field theory

a : Intermediate ( E I F ) --+ Subgroups (G), a ( K ) = Gal(E/K),

p : Subgroups (G) -4 Intermediate ( E I F ) , B(H) = Fix(H),

In particular, I GI = [ E : F], and [HI = [ E : Fix(H) ] and (G : H ) = [Fix(H) : F ] for any H

1.2 A short review of basic field theory

-+ Gal(E/(E n K ) ) , a I-, alE,

Conversely, whenever the Galois group G of a finite Galois extension

1.3 The Vahlen-Capelli Criterion

is a monic polynomial in F [ X ] , and is called the n-th cyclotomic polynomial over F.

1.2.11. ABELIAN EXTENSION,

LEMMA 1.3.2. Let p E l P and a E F . Then XP - a is irreducible in F [ X ] if and only if a @ F P .

1.3 T h e Vahlen-Capelli Criterion