Arithmetic Compactifications of PEL-Type Shimura Varieties (London Mathematical Society Monographs) - Hardcover

Lan, Kai-Wen

 
9780691156545: Arithmetic Compactifications of PEL-Type Shimura Varieties (London Mathematical Society Monographs)

Synopsis

By studying the degeneration of abelian varieties with PEL structures, this book explains the compactifications of smooth integral models of all PEL-type Shimura varieties, providing the logical foundation for several exciting recent developments. The book is designed to be accessible to graduate students who have an understanding of schemes and abelian varieties.


PEL-type Shimura varieties, which are natural generalizations of modular curves, are useful for studying the arithmetic properties of automorphic forms and automorphic representations, and they have played important roles in the development of the Langlands program. As with modular curves, it is desirable to have integral models of compactifications of PEL-type Shimura varieties that can be described in sufficient detail near the boundary. This book explains in detail the following topics about PEL-type Shimura varieties and their compactifications:



  • A construction of smooth integral models of PEL-type Shimura varieties by defining and representing moduli problems of abelian schemes with PEL structures

  • An analysis of the degeneration of abelian varieties with PEL structures into semiabelian schemes, over noetherian normal complete adic base rings

  • A construction of toroidal and minimal compactifications of smooth integral models of PEL-type Shimura varieties, with detailed descriptions of their structure near the boundary


Through these topics, the book generalizes the theory of degenerations of polarized abelian varieties and the application of that theory to the construction of toroidal and minimal compactifications of Siegel moduli schemes over the integers (as developed by Mumford, Faltings, and Chai).

"synopsis" may belong to another edition of this title.

About the Author

Kai-Wen Lan is assistant professor of mathematics at the University of Minnesota.

Excerpt. © Reprinted by permission. All rights reserved.

Arithmetic Compactifications of PEL-Type Shimura Varieties

By Kai-Wen Lan

PRINCETON UNIVERSITY PRESS

Copyright © 2013 Princeton University Press
All rights reserved.
ISBN: 978-0-691-15654-5

Contents

Acknowledgments............................................................xi
Introduction...............................................................xiii
1 Definition of Moduli Problems............................................1
2 Representability of Moduli Problems......................................91
3 Structures of Semi-Abelian Schemes.......................................143
4 Theory of Degeneration for Polarized Abelian Schemes.....................175
5 Degeneration Data for Additional Structures..............................285
6 Algebraic Constructions of Toroidal Compactifications....................373
7 Algebraic Constructions of Minimal Compactifications.....................447
A Algebraic Spaces and Algebraic Stacks....................................487
B Deformations and Artin's Criterion.......................................519
Bibliography...............................................................535
Index......................................................................545

Excerpt

<h2>CHAPTER 1</h2><p><b>Definition of Moduli Problems</b></p><br><p>In this chapter, we give the definition of the moduli problems providing integralmodels of PEL-type Shimura varieties that we will compactify.</p><p>Just to make sure that potential logical problems do not arise in our use ofcategories, we assume that a pertinent choice of a <i>universe</i> has been made (seeSection A.1.1 for more details). This is harmless for our study, and we shall notmention it again in our work.</p><p>The main objective in this chapter is to state Definition 1.4.1.4 with justifications.In order to explain the relation between our definition and those in theliterature, we include also Definition 1.4.2.1 (which, in particular, agrees with thedefinition in [83, §5] when specialized to the same bases), and compare our twodefinitions. All sections preceding them are preparatory in nature. Technical resultsworth noting are Propositions 1.1.2.20, 1.1.5.17, 1.2.2.3, 1.2.3.7, 1.2.3.11,1.2.5.15, 1.2.5.16, and 1.4.3.4. Theorem 1.4.1.11 (on the representability of ourmoduli problems in the category of algebraic stacks) is stated in Section 1.4, but itsproof will be carried out in Chapter 2. The representability of our moduli problemas schemes (when the level is neat) will be deferred until Corollary 7.2.3.10, afterwe have accomplished the construction of the minimal compactifications.</p><br><p><b>1.1 PRELIMINARIES IN ALGEBRA</p><p>1.1.1 Lattices and Orders</b></p><p>For the convenience of readers, we shall summarize certain basic definitions andimportant properties of lattices over an order in a (possibly noncommutative) finite-dimensionalalgebra over a Dedekind domain. Our main reference for this purposewill be [114].</p><p>Let us begin with the most general setting. Let <i>R</i> be a (commutative) noetherianintegral domain with fractional field Frac<i>(R)</i>.</p><p>Definition 1.1.1.1. <i>An R-lattice M is a finitely generated R-module M withno nonzero R-torsion. Namely, for every nonzero m [member of] M, there is no nonzeroelement r [member of] R such that rm = 0.</i></p><p>Note that in this case we have an embedding from <i>M</i> to <i>M</i> [??] Frac<i>(R)</i>.</p><p>Definition 1.1.1.2. <i>Let V be any finite-dimensional Frac(R)-vector space. Afull R-lattice M in V is a finitely generated submodule M of V such that Frac(R)·M = V. In other words, M contains a Frac(R)-basis of V.</i></p><p>Let <i>A</i> be a (possibly noncommutative) finite-dimensional algebra overFrac<i>(R)</i>.</p><p>Definition 1.1.1.3. <i>An R-order O in the Frac(R)-algebra A is a subring ofA having the same identity element as A, such that O is also a full R-lattice in A.</i></p><p>Here are two familiar examples of orders:</p><p>1. If <i>R</i> is a Dedekind domain, and if <i>A = L</i> is a finite separable field extensionof Frac<i>(R)</i>, then the integral closure <i>O</i> of <i>R</i> in <i>L</i> is an <i>R</i>-order in <i>A</i>. In particular, if <i>R = Z</i>, then the rings of algebraic integers <i>O = O<sub>L</sub></i> in <i>L</i> is aZ-order in <i>L</i>.</p><p>2. If <i>A</i> = M<sub><i>n</i></sub>(Frac<i>(R)</i>), then <i>O</i> = M<i><sub>n</sub>(R)</i> is an <i>R</i>-order in <i>A</i>.</p><p>Definition 1.1.1.4. <i>A maximal R-order in A is an R-order not properlycontained in another R-order in A.</i></p><p>Proposition 1.1.1.5 ([114, Thm. 8.7]). <i>1. If the integral closure of R inA is an R-order, then it is automatically maximal.</p><p>2. If O is a maximal R-order in A, then M<sub>n</sub>(O) is a maximal R-order in M<sub>n</sub>(A)for each integer n ≥ 1. In particular, if R is normal (namely, integrallyclosed in Frac(R)), then M<sub>n</sub>(R) is a maximal R-order in M<sub>n</sub>(Frac(R)).</i></p><br><p>Suppose moreover that <i>A</i> is a separable Frac<i>(R)</i>-algebra. By definition, <i>A</i> isArtinian and semisimple. For simplicity, we shall suppress the modifier <i>reduced</i>from traces and norms when talking about such algebras. By [114, Thm. 9.26], theassumption that <i>A</i> is a (finite-dimensional) separable Frac<i>(R)</i>-algebra implies thatthe (redduced) trace pairing Tr<sub><i>A</i>/Frac<i>(R)</i></sub> : <i>A x A</i> -> Frac<i>(R)</i> is <i>nondegenerate</i> (as pairings on Frac<i>(R)</i>-vector spaces).</p><p>An important invariant of an order defined by the trace pairing is the discriminant.</p><p>Definition 1.1.1.6. <i>Let t = [A</i> : Frac<i>(R)]. The discriminant</i></p><p>DDisc = Disc<sub><i>O/R</i></sub></p><p><i>is the ideal of R generated by the set of elements</p><p>{Det<sub>Frac(R)</sub>(Tr<sub>A/Frac(R)</sub>(x<sub>i</sub>x<sub>j</sub>))<sub>1≤i≤t,1≤j≤t</sub> : x<sub>1</sub>, ..., x<sub>t</sub> [member of] O}.</i></p><p>Remark 1.1.1.7. If <i>O</i> has a free <i>R</i>-basis <i>e</i><sub>1</sub>, ..., <i>e<sub>t</sub></i>, then each of the <i>t</i> elements<i>x</i><sub>1</sub>, ..., <i>x<sub>t</sub></i> can be expresseeeed as an <i>R</i>-linear combination of <i>e</i><sub>1</sub>, ..., <i>e<sub>t</sub></i>. Hencein this case Disc is generated by a single element</p><p><i>Det<sub>Frac(R)</sub>(Tr<sub>A/Frac(R)</sub>(e<sub>i</sub>e<sub>j</sub>))<sub>1≤i≤t,1≤j≤t</sub>.</i></p><p>Another important invariant is the following definition:</p><p>Definition 1.1.1.8. <i>The inverse different</i></p><p>Diff<sup>-1</sup> = Diff<sup>-1</sup><sub>O/R</sub></p><p><i>of O over R is defined by</i></p><p><i>Diff<sup>-1</sup><sub>O/R</sub> := {x [member of] A : Tr<sub>A/Frac(R)</sub>(xy) [subset] R [for all]y [member of] O}.</i></p><p>It is clear from the definition that Diff<sup>-1</sup><sub>O/R</sub> is a two-sided ideal in <i>A</i>, and thatthe formation of inverse differents is compatible with localizations.</p><p>Lemma 1.1.1.9. <i>Suppose O is locally free as an R-module. Then Diff<sup>-1</sup> islocally free as an R-module, and Tr<sub>A/Frac(R)</sub> induces a perfect pairing</i></p><p><i>Tr<sub>A/Frac(R)</sub> : O x Diff<sup>-1</sup> -> R.</p><p>Moreover, if</i> {e<sub>i</sub>}1≤i≤t <i>is any R</i><sub>1</sub>-<i>basis of O</i> [??] <i>R</i><sub>1</sub><i>for some localization R</i><sub>1</sub> <i>of R, then there exists a unique dual R</i><sub>1</sub>-<i>basis{f<sub>i</sub></i>}<sub>1≤i≤t</sub> <i>of</i> Diff<sup>-1</sup> [??] <i>R</i><sub>1</sub> <i>such that</i>Tr<sub><i>A</i>/Frac(R)</sub>(<i>e<sub>i</sub>f<sub>j</sub></i>) = δ<i><sub>ij</sub> for all</i> 1 ≤<i>i</i> ≤ <i>t and</i> 1 ≤ <i>j</i> ≤ <i>t</i>.</p><p>Proof. We may localize <i>R</i> and assume that <i>O</i> is free over <i>R</i>. Let {<i>e<sub>i</sub></i>}<sub>1≤<i>i</i>≤<i>t</i></sub>be any basis of <i>O</i> over <i>R</i>. Then {<i>e<sub>i</sub></i>}<sub>1≤<i>i</i>≤<i>t</i></sub> is also a basis of <i>A</i> over Frac<i>(R)</i>. Bynondegeneracy of the trace pairing Tr<sub><i>A</i>/Frac<i>(R)</i></sub> : <i>A x A</i> -> Frac<i>(R)</i>, there existsa unique basis <i>{f<sub>i</sub>}<sub>1≤i≤t</sub></i> of <i>A</i> over Frac<i>(R)</i>, which is dual to<i>{e<sub>i</sub>}<sub>1≤i≤t</sub></i> in thesense that Tr<sub><i>A</i>/Frac<i>R</i></sub>(<i>e<sub>i</sub>f<sub>j</sub></i>) = <i>δ<sub>ij</sub></i>for all 1 ≤ <i>i ≤ t</i> and 1 ≤ <i>j ≤ t</i>. If[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]<i>A</i> satisfies Tr<sub><i>A</i>/Frac(<i>R</i>)</sub><i>(xy)</i> [member of] <i>R</i> for all <i>x [member of] O</i>, then in particular,<i>c<sub>j</sub> = Tr<sub>A/Frac(R)</sub>(e<sub>j</sub>y) [member of] R</i> for all1 ≤ <i>j ≤ t</i>. Thus <i>{f<sub>j</sub>}1≤j≤t</i> is also a basis ofDiff<sup>-1</sup>. This shows the perfectness of the pairing Tr<sub><i>A</i>/Frac(<i>R</i>)</sub>: <i>O</i> x Diff<sup>-1</sup> -> <i>R</i>and the existence of the dual bases, as desired.</p><p>Suppose <i>R</i> is a noetherian <i>normal</i> domain, and <i>A</i> is a finite-dimensional <i>separable</i>Frac<i>(R)</i>-algebra.</p><p>Proposition 1.1.1.10 ([114, Cor. 10.4]). <i>Let R be a noetherian normal domain,and let A be a finite-dimensional separable Frac(R)-algebra. Then everyR-order in A is contained in a maximal R-order in A. There exists at least onemaximal R-order in A.</i></p><p>For each ideal p of <i>R</i>, we denote by <i>R</i><sub>p</sub> the localization of <i>R</i> at p, and by [??]<sub>p</sub>the completion of <i>R</i><sub>p</sub> with respect to its maximal ideal p<i>R</i><sub>p</sub>. (The slight deviationof this convention is that we shall denote by Z<sub>(p)</sub> the localization of Z at <i>(p)</i>, andby Z<sub><i>p</i></sub> the completion of Z<sub><i>(p)</i></sub>.) If <i>R</i> is local, then we denote simply by [??] itscompletion at its maximal ideal.</p><p>Definition 1.1.1.11. <i>Let R be a noetherian normal domain, and let p be aprime ideal of R. We say that an R-order O in A is maximal at p if O [??]R<sub>p</sub> is maximal in A.</i></p><p>Proposition 1.1.1.12 ([114, Thm. 11.1, Cor. 11.2]). <i>Let R be a noetheriannormal domain. An R-order O in A is maximal if and only if O is maximal at everyprime ideal of R, or equivalently at every maximal ideal of R.</i></p><p>Proposition 1.1.1.13 ([114, Thm. 11.5]). <i>Let R be a local noetherian normaldomain. Suppose [??] is a noetherian domain. Then an R-order O in A ismaximal if and only if O [??] [??] is an [??]-maximal order in A [??] Frac(R).</i></p><p>Remark 1.1.1.14. The statement and proof of [114, Thm. 11.5] make senseonly when [??] is a noetherian integral domain.</p><p>Corollary 1.1.1.15. <i>Let R be a noetherian normal domain, and let p bea prime ideal of R. Suppose [??]<sub>p</sub> is a noetherian domain. Then an R-order O ismaximal at p if and only if O [??] [??]<sub>p</sub> is maximal in A [??] Frac([??]<sub>p</sub>).</i></p><p>Now suppose <i>R</i> is a <i>Dedekind domain</i>. In particular, <i>R</i> is noetherian and normal,and the completions of <i>R</i> at localizations of its prime ideals are noetheriandomains.</p><p>Proposition 1.1.1.16. <i>Let</p><p>Diff = Diff<sub>O/R</sub> := {z [member of] A : z Diff<sup>-1</sup><sub>O/R</sub> [subset] O}</i></p><p><i>be the inverse ideal of</i> Diff<sup>-1</sup><sub>O/R</sub>. <i>Then this is a two-sided ideal of O, and the discriminantDisc<sub>O/R</sub> is related to Diff<sup>-1</sup><sub>O/R</sub> by</p><p>Disc<sub>O/R</sub> = Norm<sub>A/Frac(R)</sub>(Diff<sub>O/R</sub>) = [Diff<sup>-1</sup><sub>O/R</sub> : O]<sub>R</sub>.</i> (1.1.1.17)</p><p>If <i>O</i> is a maximal order, then this is just [114, Thm. 25.2]. The same proof vialocalizations works in the case where <i>O</i> is not maximal as well:</p><p>Proof of Proposition 1.1.1.16. By replacing <i>R</i> with its localizations, wemay assume that every <i>R</i>-lattice is free over <i>R</i>. Let <i>t = [A</i> : Frac<i>(R)</i>]. Let<i>{e<sub>i</sub>}<sub>1≤i≤t</sub></i> be any <i>R</i>-basis of <i>O</i>, and let{f<sub>i</sub>}<sub>1≤i≤t</sub> be the dual <i>R</i>-basis of Diff<sup>-1</sup>given by Lemma 1.1.1.9. Since <i>O</i> [subset] Diff<sup>-1</sup>, we may express each <i>e<sub>i</sub></i> as[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] for some <i>a<sub>ij</sub> [member of] R</i>. By definition,</p><p><i>Norm<sub>A/Frac(R)</sub>(Diff) = [Diff<sup>-1</sup> : O]<sub>R</sub> = (Det<sub>Frac(R)</sub>(a<sub>ij</sub>)).</i></p><p>On the other hand,</p><p>[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].</p><p>Hence Disc = <i>(Det<sub>Frac(R)</sub>(Tr<sub>A/Frac(R)</sub>(e<sub>i</sub>x<sub>j</sub>)))= (Det<sub>Frac(R)</sub>(a<sub>ij</sub>))</i>, verifyingequation (1.1.1.17) as desired.</p><p>Definition 1.1.1.18. <i>We say that the prime ideal p of R is unramified in Oif p [??] Disc<sub>O/R</sub>.</i></p><p>Proposition 1.1.1.19 ([114, Thm. 25.3]). <i>Every two maximal R-orders O<sub>1</sub>and O<sub>2</sub> in A have the same discriminant over R.</i></p><p>Therefore it makes sense to say,</p><p>Definition 1.1.1.20. <i>An ideal p of R is unramified in A if it is unramified inone (and hence every) maximal R-order of A.</i></p><p>For each global field <i>K</i>, we shall denote the rings of integers in <i>K</i> by <i>O<sub>K</sub></i>. Thisis in conflict with the notation <i>O</i> with no subscripts, but the correct interpretationshould be clear from the context.</p><p>Proposition 1.1.1.21. <i>Let R be a Dedekind domain such that Frac(R) isa global field, let A be a finite-dimensional Frac(R)-algebra with center E, andlet O be an R-order in A. Suppose p is a nonzero prime ideal of R such thatp [??] Disc<sub>O/R</sub>. Then,</p><p>1. O is maximal at p;</p><p>2. O [??] [??]<sub>p</sub> is isomorphic to a product of matrix algebras containing O<sub>E</sub> [??][??]<sub>p</sub> as its center;</p><p>3. p is unramified in A and in E.</i></p><br><p>Proof. If O [subset] O' are two orders, then necessarily</p><p><i>O [subset] O' [subset] Diff<sup>-1</sup><sub>O'/R</sub> [subset] Diff<sup>-1</sup><sub>O/R</sub>.</i></p><p>In particular, if p is a prime ideal of <i>R</i> such that p [??] Disc<sub><i>O/R</i></sub>, then the relation<i>O [??] R<sub>p</sub></i> = Diff<sup>-1</sup><sub><i>O/R</i></sub> [??] <i>R</i><sub>p</sub> forces <i>O [??] R</i><sub>p</sub>to be maximal. This proves the first statement.</p><p>According to [114, Thm. 10.5], <i>O [??] R</i><sub>p</sub> is a maximal <i>R</i><sub>p</sub>-order if and only ifit is a maximal <i>O<sub>E</sub> [??] R</i><sub>p</sub>-order. Then [114, Thm. 25.7] implies that<i>O</i> [??] [??]<sub>p</sub> is a product of matrix algebras containing <i>O<sub>E</sub></i> [??] [??]<sub>p</sub> as its center.This is the second statement.</p><p>Finally, by taking the matrix with only one element on the diagonal, we see that<i>O [??} R</i><sub>p</sub> = Diff<sup>-1</sup><i><sub>O/R</sub> [??] R</i><sub>p</sub>forces [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. That is, p [??] Disc<sub><i>O/R</i></sub>forces [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. Then the third statement follows from Proposition 1.1.1.19and Definition 1.1.1.20.</p><p>Definition 1.1.1.22. <i>A (left) O-module M is called an O-lattice if it is anR-lattice. Namely, it is finitely generated and torsion-free as an R-module.</i></p><p>Proposition 1.1.1.23 (see [114, Thm. 21.4 and Cor. 21.5]). <i>Every maximalorder O over a Dedekind domain is hereditary in the sense that all O-lattices areprojective O-modules.</i></p><p>Proposition 1.1.1.24 (see [114, Cor. 21.5 and Thm. 2.44]). <i>Every projectivemodule over a maximal order is a direct sum of left ideals.</i></p><p>Remark 1.1.1.25. Propositions 1.1.1.23 and 1.1.1.24 imply that, althoughO-lattices might not be projective, their localizations or completions become projectiveas soon as O itself becomes maximal after localization or completion.</p><br><p><b>1.1.2 Determinantal Conditions</b></p><p>Let C be a finite-dimensional <i>separable</i> algebra over a field <i>k</i>. By definition (suchas [114, p. 99]), the center <i>E</i> of <i>C</i> is a commutative finite-dimensional separablealgebra over <i>k</i>. Let <i>K</i> be a (possibly infinite) field extension of <i>k</i>. Unless otherwisespecified, all the homomorphisms below will be <i>k</i>-linear.</p><p>Fix a separable closure <i>K</i><sup>sep</sup> of <i>K</i>, and consider the possible <i>k</i>-algebra homomorphismsτ from <i>E</i> to <i>K</i><sup>sep</sup>. Note that Hom<i><sub>k</sub>(E,K</i><sup>sep</sup>) has cardinality <i>[E : k]</i>,because E is separable over k. The Gal(Ksep/K)-orbits [t ] of such homomorphisms<i>τ : E -> K</i><sup>sep</sup> can be classified in the following way: Consider the equivalenceclasses of pairs of the form (<i>K</i><sub>τ</sub>, τ), where <i>K</i>τ is isomorphic over <i>K</i> to thecomposite of <i>K</i> and the image of τ in <i>K</i><sup>sep</sup>, and where τ is the induced homomorphismfrom <i>E</i> to <i>K</i><sub>τ</sub>. Here <i>K</i><sub>τ</sub> is necessarily separable over <i>K</i> with degree atmost [<i>E : k</i>]. Two such pairs [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]and [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] are considered equivalent ifthere is an isomorphism [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]over <i>K</i> such that τ<sub>2</sub> = σ [??] τ<sub>1</sub>. We shalldenote such an equivalence class by [τ] : <i>E -> K<sub>[τ]</sub></i>. By abuse of notation, thiswill also mean an actual representative τ : <i>E -> K<sub>τ</sub></i>, which can be considered as ahomomorphism τ : <i>E -> K</i><sup>sep</sup> as well. Note that</p><p>[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]</p><p>where each K<sup>sep</sup><sub>τ'</sub> means a copy of <i>K</i><sup>sep</sup> withτ' : <i>E -> K</i><sup>sep</sup> in the equivalenceclass [τ].</p><p>Lemma 1.1.2.1. <i>We have a decomposition</i></p><p>[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]</p><p><i>into a product of separable extensions E [τ] of K.</i></p><p>Corollary 1.1.2.2. <i>We have a decomposition</i></p><p>[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]</p><p><i>into simple K-algebras.</i></p><p>Let us quote the following weaker form of the Noether–Skolem theorem:</p><p>LEMMA 1.1.2.4 (see, for example, [68, Lem. 4.3.2]). <i>Let C' be a simple Artinianalgebra. Then all simple C'-modules are isomorphic to each other.</i></p><p>A useful reformulation of Lemma 1.1.2.4 is as follows:</p><p>Corollary 1.1.2.5. <i>Let C' be a simple Artinian algebra with center E'. Thenan irreducible representation of C' with coefficients in some field K' is determinedup to isomorphism by its restriction to E'.</i></p><p><i>(Continues...)</i>

(Continues...)
Excerpted from Arithmetic Compactifications of PEL-Type Shimura Varieties by Kai-Wen Lan. Copyright © 2013 by Princeton University Press. Excerpted by permission of PRINCETON UNIVERSITY PRESS.
All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.

"About this title" may belong to another edition of this title.