AN INTRODUCTION TO O-MINIMAL GEOMETRY. Michel COSTE Institut de Recherche Mathématique de Rennes - PDF

Description
AN INTRODUCTION TO O-MINIMAL GEOMETRY Michel COSTE Institut de Recherche Mathématique de Rennes November 1999 Preface These notes have served as a basis for a course in Pisa

Please download to get full document.

View again

of 22
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Information
Category:

Film

Publish on:

Views: 24 | Pages: 22

Extension: PDF | Download: 0

Share
Transcript
AN INTRODUCTION TO O-MINIMAL GEOMETRY Michel COSTE Institut de Recherche Mathématique de Rennes November 1999 Preface These notes have served as a basis for a course in Pisa in Spring A parallel course on the construction of o-minimal structures was given by A. Macintyre. The content of these notes owes a great deal to the excellent book by L. van den Dries [vd]. Some interesting topics contained in this book are not included here, such as the Vapnik-Chervonenkis property. Part of the material which does not come from [vd] is taken from the paper [Co1]. This includes the sections on the choice of good coordinates and the triangulation of functions in Chapter 4 and Chapter 5. The latter chapter contains the results on triviality in families of sets or functions which were the main aim of this course. The last chapter on smoothness was intended to establish property DC k - all k which played a crucial role in the course of Macintyre (it can be easily deduced from the results in [vdmi]). It is also the occasion to give a few results on tubular neighborhoods. I am pleased to thank Francesca Acquistapace, Fabrizio Broglia and all colleagues of the Dipartimento di Matematica for the invitation to give this course in Pisa and their friendly hospitality. Also many thanks to Antonio Diaz-Cano, Pietro Di Martino, Jesus Escribano and Federico Ponchio for reading these notes and correcting mistakes. 1 2 Contents 1 O-minimal Structures Introduction Semialgebraic Sets Definition of an O-minimal Structure Cell Decomposition Monotonicity Theorem Cell Decomposition Connected Components and Dimension Curve Selection Connected Components Dimension Definable Triangulation Good Coordinates Simplicial Complex Triangulation of Definable Sets Triangulation of Definable Functions Generic Fibers for Definable Families The Program The Space of Ultrafilters of Definable Sets The O-minimal Structure κ(α) Extension of Definable Sets Definable Families of Maps Fiberwise and Global Properties Triviality Theorems Topological Types of Sets and Functions 4 CONTENTS 6 Smoothness Definable Functions in One Variable C k Cell Decomposition Definable Manifolds and Tubular Neighborhoods Bibliography 79 Index 81 Chapter 1 O-minimal Structures 1.1 Introduction The main feature of o-minimal structures is that there are no monsters in such structures. Let us take an example of the pathological behaviour that is ruled out. Let Γ R 2 be the graph of the function x sin(1/x) for x 0, and let clos(γ) be the closure of Γ in R 2. This set clos(γ) is the union of Γ and the closed segment joining the two points (0, 1) and (0, 1). We have dim(clos(γ) \ Γ)=dimΓ=1. Observe also that clos(γ) is connected, but not arcwise connected: there is no continuous path inside clos(γ) joining the origin with a point of Γ. The o-minimal structures will allow to develop a tame topology in which such bad things cannot happen. The model for o-minimal structures is the class of semialgebraic sets. A semialgebraic subset of R n is a subset defined by a boolean combination of polynomial equations and inequalities with real coefficients. In other words, the semialgebraic subsets of R n form the smallest class SA n of subsets of R n such that: 1. If P R[X 1,...,X n ], then {x R n ; P (x) =0} SA n and {x R n ; P (x) 0} SA n. 2. If A SA n and B SA n, then A B, A B and R n \ A are in SA n. On the one hand the class of semialgebraic sets is stable under many constructions (such as taking projections, closure, connected components...),and on the other hand the topology of semialgebraic sets is very simple, without 5 6 CHAPTER 1. O-MINIMAL STRUCTURES pathological behaviour. The o-minimal structures may be seen as an axiomatic treatment of semialgebraic geometry. An o-minimal structure (expanding the field of reals) is the data, for every positive integer n, of a subset S n of the powerset of R n, satisfying certain axioms. There are axioms which allow to perform many constructions inside the structure, and an o-minimality axiom which guarantee the tameness of the topology. One can distinguish two kinds of activities in the study of o-minimal structures. The first one is to develop the geometry of o-minimal structures from the axioms. In this activity one tries to follow the semialgebraic model. The second activity is to discover new interesting classes statisfying the axioms of o-minimal structures. This activity is more innovative, and the progress made in this direction (starting with the proof by Wilkie that the field of reals with the exponential function defines an o-minimal structure) justifies the study of o-minimal structures. The subject of this course is the first activity (geometry of o-minimal structures), while the course of A. Macintyre is concerned with the construction of o-minimal structures. 1.2 Semialgebraic Sets The semialgebraic subsets of R n were defined above. We denote by SA n the set of all semialgebraic subsets of R n. Some stability properties of the class of semialgebraic sets follow immediately from the definition. 1. All algebraic subsets of R n are in SA n. Recall that an algebraic subset is a subset defined by a finite number of polynomial equations P 1 (x 1,...,x n )=...= P k (x 1,...,x n )=0. 2. SA n is stable under the boolean operations, i.e. finite unions and intersections and taking complement. In other words, SA n is a Boolean subalgebra of the powerset P(R n ). 3. The cartesian product of semialgebraic sets is semialgebraic. If A SA n and B SA p, then A B SA n+p. Sets are not sufficient, we need also maps. Let A R n be a semialgebraic set. A map f : A R p is said to be semialgebraic if its graph Γ(f) R n R p = R n+p is semialgebraic. For instance, the polynomial maps are semialgebraic. The function x 1 x 2 for x 1 is semialgebraic. 1.2. SEMIALGEBRAIC SETS 7 The most important stability property of semialgebraic sets is known as Tarski-Seidenberg theorem. This central result in semialgebraic geometry is not obvious from the definition. 4. Denote by p : R n+1 R n the projection on the first n coordinates. Let A be a semialgebraic subset of R n+1. Then the projection p(a) is semialgebraic. The Tarski-Seidenberg theorem has many consequences. For instance, it implies that the composition of two semialgebraic maps is semialgebraic. We shall say more on the consequences of the stability under projection in the context of o-minimal structures. The semialgebraic subsets of the line are very simple to describe. Indeed, a semialgebraic subset A of R is described by a boolean combination of sign conditions ( 0, = 0 or 0) on polynomials in one variable. Consider the finitely many roots of all polynomials appearing in the definition of A. The signs of the polynomials are constant on the intervals delimited by these roots. Hence, such an interval is either disjoint from or contained in A. We obtain the following description. 5. The elements of SA 1 are the finite unions of points and open intervals. We cannot hope for such a simple description of semialgebraic subsets of R n, n 1. However, we have that every semialgebraic set has a finite partition into semialgebraic subsets homeomorphic to open boxes (i.e. cartesian product of open intervals). This is a consequence of the so-called cylindrical algebraic decomposition (cad), which is the main tool in the study of semialgebraic sets. Actually, the Tarki-Seidenberg theorem can be proved by using cad. A cad of R n is a partition of R n into finitely many semialgebraic subsets (which are called the cells of the cad), satisfying certain properties. We define the cad of R n by induction on n. A cad of R is a subdivision by finitely many points a 1 ... a l. The cells are the singletons {a i } and the open intervals delimited by these points. For n 1, a cad of R n is given by a cad of R n 1 and, for each cell C of R n 1, continuous semialgebraic functions ζ C,1 ... ζ C,lC : C R. 8 CHAPTER 1. O-MINIMAL STRUCTURES The cells of R n are the graphs of the functions ζ C,i and the bands in the cylinder C R R n delimited by these graphs. For i =0,...,l C, the band (ζ C,i,ζ C,i+1 )is (ζ C,i,ζ C,i+1 )={(x,x n ) R n ; x C and ζ C,i (x ) x n ζ C,i+1 (x )}, where we set ζ C,0 = and ζ C,lC +1 =+. Observe that every cell of a cad is homeomorphic to an open box. This is proved by induction on n, since a graph of ζ C,i is homeomorphic to C and a band (ζ C,i,ζ C,i+1 ) is homeomorphic to C (0, 1). Given a finite list (P 1,...,P k ) of polynomials in R[X 1,...,X n ], a subset A of R n is said to be (P 1,...,P k )-invariant if the sign ( 0, = 0 or 0) of each P i is constant on A. The main result concerning cad is the following. Theorem 1.1 Given a finite list (P 1,...,P k ) of polynomials in R[X 1,...,X n ], there is a cad of R n such that each cell is (P 1,...,P k )-invariant. It follows from Theorem 1.1 that, for every semialgebraic subset A of R n, there is a cad such that A is a union of cells (such a cad is called adapted to A). Indeed, A is defined by a boolean combination of sign conditions on a finite list of polynomials (P 1,...,P k ), and it suffices to take a cad such that each cell is (P 1,...,P k )-invariant. We illustrate Theorem 1.1 with the example of a cad of R 3 such that each cell is (X X X 2 3 1)-invariant (a cad adapted to the unit sphere). Such a cad is shown on Figure 1.1. Exercise 1.2 How many cells of R 3 are there in this cad? Is it possible to have a cad of R 3, adapted to the unit sphere, with less cells? We refer to [Co2] for an introduction to semialgebraic geometry. Semialgebraic geometry can also be developed over an arbitrary real closed field, instead of the field of reals. A real closed field R is an ordered field satisfying one of the equivalent conditions: Every positive element is a square and every polynomial in R[X] with odd degree has a root in R. For every polynomial F R[X] and all a, b in R such that a band F (a)f (b) 0, there exists c R, a c b, such that F (c) =0. R[ 1] = R[X]/(X 2 + 1) is an algebraically closed field. 1.3. DEFINITION OF AN O-MINIMAL STRUCTURE 9 Figure 1.1: A cad adapted to the sphere A semialgebraic subset of R n is defined as for R n. The properties of semialgebraic sets that we have recalled in this section also hold for semialgebraic subsets of R n. We refer to [BCR] for the study of semialgebraic geometry over an arbitrary real closed field. 1.3 Definition of an O-minimal Structure We shall now define the o-minimal structures expanding a real closed fied R. The fact that we consider a situation more general than the field of reals will be important only in Chapter 5. Otherwise, the reader may take R = R. An interval in R will always be an open interval (a, b) (for a b)or(a, + ) or (,b). We insist that an interval always has endpoints in R {, + }. For instance, if R is the field of real algebraic numbers (which is the smallest real closed field), the set of x in R such that 0 x π(π = ) is not an interval, because there is no right endpoint in R. Occasionally, we shall also use the notation [a, b] for closed segments in R. The field R has a topology for which the intervals form a basis. Affine spaces R n are endowed with the product topology. The open boxes, i.e. the 10 CHAPTER 1. O-MINIMAL STRUCTURES cartesian products of open intervals (a 1,b 1 ) (a n,b n ) form a basis for this topology. The polynomials are continuous for this topology. Exercise 1.3 Prove that the polynomials are continuous. When dealing with the topology, one should take into account that a real closed field is generally not locally connected nor locally compact (think of real algebraic numbers). Definition 1.4 A structure expanding the real closed field R is a collection S =(S n ) n N, where each S n is a set of subsets of the affine space R n, satisfying the following axioms: 1. All algebraic subsets of R n are in S n. 2. For every n, S n is a Boolean subalgebra of the powerset of R n. 3. If A S m and B S n, then A B S m+n. 4. If p : R n+1 R n is the projection on the first n coordinates and A S n+1, then p(a) S n. The elements of S n are called the definable subsets of R n. The structure S is said to be o-minimal if, moreover, it satisfies: 5. The elements of S 1 are precisely the finite unions of points and intervals. In the following, we shall always work in a o-minimal structure expanding a real closed field R. Definition 1.5 A map f : A R p (where A R n ) is called definable if its graph is a definable subset of R n R p. (Applying p times property 4, we deduce that A is definable). Proposition 1.6 The image of a definable set by a definable map is definable. Proof. Let f : A R p be definable, where A R n, and let B be a definable subset of A. Denote by Γ f = {(x, f(x)) ; x A} R n+p the graph of f. Let be the algebraic (in fact, linear) subset of R p+n+p consisting of those (z, x, y) R p R n R p such that z = y. Then C = (R p Γ) (R p B R p ) is a definable subset of R p+n+p. Let p p+n+p,p : R p+n+p R p be the projection on the first p coordinates. We have p p+n+p,p (C) =f(b), and, applying n + p times property 4, we deduce that f(b) is definable. Observe that every polynomial map is a definable map, since its graph is an algebraic set. 1.3. DEFINITION OF AN O-MINIMAL STRUCTURE 11 Exercise 1.7 Every semialgebraic subset of R n is definable (Hint: the set defined by P (x 1,...,x n ) 0 is the projection of the algebraic set with equation x 2 n+1p (x 1,...,x n ) 1 = 0). Hence, the collection of SA n is the smallest o-minimal structure expanding R. Exercise 1.8 Show that every nonempty definable subset of R has a least upper bound in R {+ }. Exercise 1.9 Assume that S is an o-minimal structure expanding an ordered field R (same definition as above). Show that R is real closed. (Hint: one can use the second equivalent condition for a field to be real closed, the continuity of polynomials and property 5 of o-minimal structures.) Exercise 1.10 Let f =(f 1,...,f p ) be a map from A R n into R p. Show that f is definable if and only if each of its coordinate functions f 1,...,f p is definable Exercise 1.11 Show that the composition of two definable maps is definable. Show that the definable functions A R form an R-algebra. Proposition 1.12 The closure and the interior of a definable subset of R n are definable. Proof. It is sufficient to prove the assertion concerning the closure. The case of the interior follows by taking complement. Let A be a definable subset of R n. The closure of A is clos(a) = { x R n ; ε R, ε 0 y R n,y A and } n (x i y i ) 2 ε 2 i=1. where x =(x 1,...,x n ) and y =(y 1,...,y n ). The closure of A can also be described as clos(a) =R n \ ( p n+1,n ( R n+1 \ p 2n+1,n+1 (B) )), where { B =(R n R A) (x, ε, y) R n R R n ; } n (x i y i ) 2 ε 2, i=1 12 CHAPTER 1. O-MINIMAL STRUCTURES p n+1,n (x, ε) =x and p 2n+1,n+1 (x, ε, y) =(x, ε). Then observe that B is definable. The example above shows that it is usually boring to write down projections in order to show that a subset is definable. We are more used to write down formulas. Let us make precise what is meant by a first-order formula (of the language of the o-minimal structure). A first-order formula is constructed according to the following rules. 1. If P R[X 1,...,X n ], then P (x 1,...,x n )=0andP (x 1,...,x n ) 0 are first-order formulas. 2. If A is a definable subset of R n, then x A (where x =(x 1,...,x n )) is a first-order formula. 3. If Φ(x 1,...,x n ) and Ψ(x 1,...,x n ) are first-order formulas, then Φ and Ψ, Φ or Ψ, not Φ, Φ Ψ are first order formulas. 4. If Φ(y, x) is a first-order formula (where y = (y 1,...,y p ) and x = (x 1,...,x p )) and A is a definable subset of R n, then x A Φ(y, x) and x A Φ(y, x) are first-order formulas. Theorem 1.13 If Φ(x 1,...,x n ) is a first-order formula, the set of (x 1,...,x n ) in R n which satisfy Φ(x 1,...,x n ) is definable. Proof. By induction on the construction of formulas. Rule 1 produces semialgebraic sets, which are definable. Rule 2 obviously produces definable sets. Rule 3 works because SA n is closed under boolean operation. Rule 4 reflects the fact that the projection of a definable set is definable. Indeed, if B = {(y, x) R p+n ;Φ(y, x)} is definable and p p+n,p : R p+n R p denotes the projection on the first p coordinates, we have {y R p ; x A Φ(y, x)} = p p+n,p ((R p A) B), {y R p ; x A Φ(y, x)} = R p \ p p+n,p ((R p A) (R p+n \ B)), which shows that both sets are definable. One should pay attention to the fact that the quantified variables have to range over definable sets. For instance, {(x, y) R 2 ; n N y = nx} is not definable (why?). We refer the reader to [Pr] for notions of model theory. 1.3. DEFINITION OF AN O-MINIMAL STRUCTURE 13 Exercise 1.14 Let f : A R be a definable function. Show that {x A ; f(x) 0} is definable. Hence, we can accept inequalities involving definable functions in formulas defining definable sets. Exercise 1.15 Let A be a non empty definable subset of R n. For x R n, define dist(x, A) as the greatest lower bound of the set of y x = ni=1 (y i x i ) 2 for y A. Show that dist(x, A) is well-defined and that x dist(x, A) is a continuous definable function on R n. Exercise 1.16 Let f : A R be a definable function and a clos(a). For ε 0, define m(ε) = inf{f(x) ; x A and x a ε} R { }. Show that m is a definable function (this means that m 1 ( ) is definable and m m 1 (R) is definable). Show that lim inf x a f(x) exists in R {, + }. 14 CHAPTER 1. O-MINIMAL STRUCTURES Chapter 2 Cell Decomposition In this chapter we prove the cell decomposition for definable sets, which generalizes the cylindrical algebraic decomposition of semialgebraic sets. This result is the most important for the study of o-minimal geometry. The proofs are rather technical. The main results come from [PS, KPS], and we follow [vd] rather closely. 2.1 Monotonicity Theorem Theorem 2.1 (Monotonicity Theorem) Let f :(a, b) R be a definable function. There exists a finite subdivision a = a 0 a 1 ... a k = b such that, on each interval (a i,a i+1 ), f is continuous and either constant or strictly monotone. The key of the proof of the Monotonicity Theorem is the following Lemma. Lemma 2.2 Let f :(a, b) R be a definable function. There exists a subinterval of (a, b) on which f is constant, or there exists a subinterval on which f is strictly monotone and continuous. Proof. Suppose that there is no subinterval of (a, b) on which f is constant. First step: there exists a subinterval on which f is injective. It follows from the assumption that, for all y in R, the definable set f 1 (y) is finite: otherwise, f 1 (y) would contain an interval, on which f = y. Hence the definable set f((a, b)) is infinite and contains an interval J. The function g : J (a, b) defined by g(y) = min(f 1 (y)) is definable and satisfies f g =Id J. Since g is injective, g(j) is infinite and contains a subinterval I of (a, b). We have g f I =Id I, and f is injective on I. 15 16 CHAPTER 2. CELL DECOMPOSITION Second step: there exists a subinterval on which f is strictly monotone. We know that f is injective on a subinterval I of (a, b). Take x I. The sets I + = {y I ; f(y) f(x)} I = {y I ; f(y) f(x)} form a definable partition of I \{x}. Therefore, there is ε 0 such that (x ε, x) (resp. (x, x+ε)) is contained in I + or in I. There are four possibilities Φ +,+ (x), Φ +, (x), Φ,+ (x) and Φ, (x) which give a definable partition of I. For instance, Φ +, (x) is ε y I ( (x ε y x f(y) f(x)) and (x y x+ ε f(y) f(x)) ). We claim that Φ +,+ and Φ, are finite. It is sufficient to prove the claim for Φ +,+ (for Φ,, replace f with f). If Φ +,+ is not finite, it holds on a subinterval of I, which we still call I. Set B = {x I ; y I y x f(y) f(x)}. If B contains an interval, then f is strictly increasing on this interval, which contradicts Φ +,+. Hence, B is finite. Replacing I with (max(b), + ) I if B, we can assume ( ) x I y I (y x a
Related Search
Similar documents
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks