Mathematical World « Volume 2 Fixed Points Yu. A. Shashkin Translated from the Russian by Viktor Minachin Mathoustical Association of yee @10. A. IIADIKMH HETIOJBYDKHbIE TOUKN «HAYKA», MOCKBA, 1989 Translated from the Russian by Viktor Minachin 1991 Mathematics Subject Classification. Primary 01-01, 54-01, 54H25; Secondary 54C15. Library of Congress Cataloging-In-Pubilcation Data Shashkin, 1U. A. ((Uril Alekseevich) [Nepodvizhnye tochki. English] Fixed points/Yu. A. Shashkin; translated from the Russian by Viktor Minachin. P. cm.—(Mathematical world, ISSN 1055-9426; v. 2) Includes bibliographical references. ISBN 0-8218-9000-X (alk. paper) 1. Fixed point theory. |. Title. II, Series. QA329,9.85313 1991 91-28904 515'.7248—de20 cr Copying and reprinting. Individual readers of this publication, and nonprofit libraries acting for them, are permitted to make fair use of the material, such as to copy an article for use in teaching or research. Permission is granted to quote brief passages from this publication in reviews, provided the customary acknowledgment of the source is given. Republication, systematic copying, or multiple reproduction of any material in this pub- lication (including abstracts) is permitted only under license from the American Mathematical Society. Requests for such permission should be addressed to the Manager of Editorial Services, Amencan Mathematical Society, P.O, Box 6248, Providence, Rhode Island 02940-6248. The owner consents to copying beyond that permitted by Sections 107 or 108 of the U.S. Copyright Law, provided that a fee of $1.00 plus $.25 per page for each copy be paid directly to the Copyright Clearance Center, Inc., 27 Congress Street, Salem, Massachusetts 01970. This consent does not extend to other kinds of copying, such as copying for general distribution, for advertising or promotional purposes, for creating new collective works, or for resale. Copyright ©1991 by the American Mathematical Society. All rights reserved. Printed in the United States of America The American Mathematical Society retains all rights except those granted to the United States Government. The paper used in this book is acid-free and falls within the guidelines established to ensure permanence and durability. © This publication was typeset using AV4S-TEX, the American Mathematical Society's TEX macro system. 10987654321 96 95 95 94939291Table of Contents Preface Chapter | Chapter 2 Chapter 3 Chapter 4 Chapter 5 Chapter 6 Chapter 7 Chapter 8 Chapter 9 Chapter \0 Chapter \1 Chapter 12 Chapter 13 Chapter 14 ‘Solutions and Answers References Continuous Mappings of a Closed Interval and se eeepc cece asec eeseeeeeeeeeeeceeeeeeeeeegeaeegreeeeeeeeaeen First Combinatorial Lemma ............. Second Combinatorial Lemma, or Walks through the Rooms in a House ... i Sperner’s Lemma . . Continuous Mappings, Homeomorphisms, and the Fixed Point Property..............5- Compactness.. 1.2... 0... eee ee eee eee Proof of Brouwer’s Theorem for a Closed Inter- val, the Intermediate Value Theorem, and Ap- plications : Proof of Brouwer’s Theorem for a Square. . . The Iteration Method ..............00 0s Retraction. ...... 60.6. cece cece eee eee Continuous Mappings of a Circle, Homotopy, and Degree of a Mapping ............... Second Definition of the Degree of a Mapping Continuous Mappings of a Sphere ........ Lemma on Equality of Degrees........... vii 15 21 25 33 39 43 47 53 55 61 65Preface Applying mathematics means, in many cases, solving equations. If that is the case, then the important thing to know is whether a particular equation has a solution or not. The presence of solutions is guaranteed by so-called existence theorems. Let f be a function of the real variable x , continuous in the closed interval [a, b] and assuming values of different signs at its endpoints. Then the equation (0.1) f(x) =0 has at least one solution inside the interval. Existence theorems are often expressed in the form of “fixed point” prin- ciples. For example, let us view equation (0.1) in the following way. Write (0.1) in the form Af(x)+x =x, where A is a positive parameter. Denoting Af(x) +x by F(x) we get the equation (0.2) F(x) =x. Choose the value of 4 in such a way that all the values of F lie inside the interval [a, 6]. Equation (0.2) can now be looked upon as follows. The function F maps the point (the real number) x from the interval [a,5] into the point F(x) = y of the same interval, which, in general, does not coincide with x. In other words, the mapping F takes the point x into the point y. However, if a point x9 is a solution of (0.2), then it stays where it was; that is, it is a fixed point. The same point is evidently a solution of equation (0.1) as well. Therefore, in geometrical terms, a theorem ensuring the existence of a solution for equation (0.2) is formulated as the following fixed point principle: if F is a continuous function mapping a closed interval into itself, then the function has at least one fixed point. Because F is an arbitrary continuous function, this property is actually that of the closed interval itself; it does not depend on the choice of a particular mapping and is called the fixed point property. The problem of solving the system of two equations S(,y)=a, g(x, y)=b for unknown x and y can be reduced to the problem of whether a mapping of a square or, say, a disk into itself has a fixed point. viiviii PREFACE Fixed point theorems have numerous applications in mathematics. Most of the theorems ensuring the existence of solutions for differential, integral, operator, or other equations can be reduced to fixed point theorems. They are also used in new areas of mathematical applications, e.g., in mathematical economics, game theory, etc. The subject of this book is essentially one single problem: whether a closed interval, square, disc, or sphere has the fixed point property. The theory of fixed points belongs to topology (see [3] and [5]), a part of mathematics created at the end of the nineteenth century, and makes exten- sive use of such topological notions as continuity, compactness, homotopy, and the degree of a mapping. Another aim of the book is to show how combinatorial considerations related to decomposition (triangulation) of figures into distinct parts called faces (simplexes) adjoining each other in a regular fashion are used in this theory. Three names should be mentioned here. The first is the famous French mathematician H. Poincaré (1854-1912), the founder of the fixed point approach, who had deep insight into its future importance for problems of mathematical analysis and celestial mechanics, and took an active role in its development. Poincaré was the first to apply the combinatorial approach to topology, using triangulations of geometrical figures into simplexes. The second is the Dutch mathematician L.E.Y. Brouwer (1881-1966). He introduced the topological notions used in this book, among them those of homotopy and the degree of a mapping. He also proved the fixed point theorems for a square, a sphere, and their n-dimensional counterparts. The third is the German mathematician E. Sperner (1906-1980), who back in 1928 proved the combinatorial geometric lemma on the decomposition of a triangle (as well as of any n-dimensional simplex in general), which plays an important role in the theory of fixed points. The author would like to express his deep gratitude to V. G. Boltyansky, whose suggestions led to substantial improvements of the first version of the manuscript. He is also indebted to E. G. Pytkeyev for useful discussions during the preparation of this book.] Continuous Mappings of a Closed Interval and a Square Consider a closed interval / = [0,1] of the real line R. Let f bea function mapping J onto itself. This means that a rule is specified associating to each point x in J a unique point f(x) of the same interval. We say that JS (x) is the image of the point x or that the point x goes into the point JS(x). Along with the term “function” we will also use the term “mapping”; for us they are synonyms. The only condition the mapping f should satisfy is that it be continuous. The meaning of continuity is intuitively clear: the graph of the mapping is a continuous curve and can be drawn without lifting a pen off the paper. Another way to describe a continuous mapping is to say that it takes close points into close ones. For the moment we will be satisfied with the general idea and postpone a precise definition until Chapter 5. Consider a square Q lying on the coordinate plane R? so that the coor- dinates of its points satisfy the following inequalities: 0< x <1, 0 lh < > > 4 4 4 Ficure 5 78g 3. SECOND COMBINATORIAL LEMMA the walk is finished, we start another one, and continue until there are no more outside doors or dead ends to start from. The resulting paths are of one of three types: (1) from an outside door into a dead end (or the other way round, which for our purposes makes no difference); (2) from an outside door to another outside door; (3) from a dead end into a dead end. Denote by m, n, and p the numbers of paths for the first, second, and third type, respectively. Because each path of type (1) corresponds to one outside door, and each path of type (2) corresponds to two outside doors, the total number of outside doors is m + 2n. Similarly, the number of dead ends is m + 2p. The numbers m+ 2n and m + 2p are of the same parity, which proves the lemma. Note that the shapes of the room are irrelevant. They can, for example, be triangular, and this is what we shall assume next.4 Sperner’s Lemma Consider an arbitrary triangle subdivided into smaller triangles. We will always assume that the subdivision satisfies the following condition: any pair of smaller triangles either have no common points, or have a common vertex, or have a common side. Such a subdivision is called a triangulation, smaller triangles are said to be the faces of the triangulation, the sides of smaller triangles are its edges, and their vertices are the vertices of the triangulation. For example, the subdivision shown in Figure 6a is a triangulation, while that in Figure 6b is not. Lemma 3 (Sperner). Consider a triangulation of a triangle T . The vertices of this triangle are labeled by 1, 2, and 3. The vertices of the triangulation are labeled by the same numbers in such a way that the following boundary condition is satisfied: if a vertex lies on a side of the triangle T,, it is labeled by one of the two numbers labeling the endpoints of this side. Then there is at least one face in the triangulation with the vertices labeled by different numbers, i.e., 1, 2, and 3. Moreover, the number of such faces is odd. Proof. An illustration to Lemma 3 is given in Figure 7 on p. 10. It is proved by reduction to Lemma 2. Let us call triangle T a house and each face of the triangulation a room. An edge of the triangulation is called a door provided its endpoints are labeled by 1 and 2. We shall say that such edges are of type (1, 2). (Note that, in contrast to Lemma 1, we make no difference Ficure 6 910 4, SPERNER’S LEMMA FicureE 7 between edges of type (1, 2) and (2, 1).) Which faces are now the dead ends? Consider all possible assignments of labels 1, 2, and 3 to the vertices of a face. They can be of ten different types: (1, 1, 1), (1, 1, 2), (1, 1, 3), (1, 2, 2), (1, 2, 3), (1, 3, 3), (2, 2, 2), (2, 2, 3), (2, 3, 3), (3, 3, 3). Evidently, only the faces of type (1, 2, 3) have exactly one edge of the type (1, 2), so it is quite natural to call them dead ends. Similarly, the faces of types (1, 1, 2) and (1, 2, 2) should be called communicating rooms, because they contain exactly two edges of the type (1, 2). We, therefore, obtain the following “glossary” for translating conditions of Lemma 3 into the language of Lemma 2: triangle T—house face of triangulation—room edge of triangulation of type (1, 2)—door boundary (resp., inside) edge of triangle T—outside (resp., internal) door face of type (1, 2, 3)—dead end face of type (1, 1, 2), or (1, 2, 2)—-communicating room Whatever the choice of labels for the vertices of a triangulation is, the list shows that a face may have one or two edges of type (1, 2), or none at all. Hence, the conditions of Lemma 2 are satisfied. Therefore, the number of dead ends and outside doors is either both even or both odd. Translating this back into the language of Lemma 3, we conclude that the number of faces of type (1, 2, 3) and the number of boundary edges of type (1, 2) are of the same parity. It remains to show that the second number is odd. But the boundary edges of type (1, 2) can lie only on the side of the triangle T labeled by 1 and 2. Because there are no edges of this type on the other sides of the triangle, the required conclusion follows from Lemma 1. Lemma 3 is proved. The process used in the proof will be called the walking process. Below we shall use it by “walking” the faces of a triangulation itself, without necessarily calling them rooms. The crucial point will always be what is meant by a “door” edge, and by a “dead end” face.4, SPERNER'S LEMMA " 4 3 3 4.3 34 cl ! Tz 5-2 arr ' aftr f2 3 5 7 rT i2 ft 2 2 7 gee) ares tear | 22: : + Hf 4 1 1 2 t ft 21 1° 2F 1 172 a b FiGure 8 Let us use this approach to prove Lemma 4, which makes the same state- ment as Lemma 3, but for a square. Let a square Q be divided into smaller squares by lines parallel to its sides. For this subdivision we will use the same names as before, i.e., faces, edges, and vertices of a subdivision, where, for example, faces are smaller squares of the subdivision. Lemma 4. Let a square Q be subdivided into smaller squares by lines paral- lel to its sides. The vertices of the square are labeled by numbers |, 2, 3, and 4. The vertices of the subdivision are labeled by the same numbers in such a way that the following boundary condition is satisfied: if a vertex of the subdivision lies on the side of the square Q, then it is labeled by one of the two numbers labeling the endpoints of this side (see Figure 8a). Then there is at least one Jace labeled by at least three different numbers. Proor. To prove the theorem let us subdivide each square face into two triangular faces (see Figure 8b). The result is a triangulation of the square, all vertices of which are labeled. Let us call an edge with the labels (1, 2) a door. A dead end is a triangular face labeled by three different numbers such that at least two of them are | and 2. Thus, dead ends are triangles of type (1, 2, 3) or (1, 2, 4). By the hypothesis of Lemma 4, all the outside doors lie on side (1, 2) of the square, and by Lemma | their number is odd. Any walk through the triangular faces of the square starts at an outside door and ends in either of the two cases: leaving the square through another outside door or entering a dead end (see Figure 8b). Since the total number of outside doors is odd, at least one of the walks must end in a dead end. This means that there is a triangular face labeled either by (1, 2, 3), or (1, 2, 4), and, consequently, there is a square face labeled by at least three different numbers. Problems. 5. Prove the first combinatorial lemma by a walking process. Consider a closed interval as a house in Lineland. What are its rooms, doors, and dead ends?12 4 SPERNER’S LEMMA Ficure 9 6. A triangulation of square Q is given, as in Figure 8b, or Figure 9. Each of the vertices of the triangulation is labeled by one of the following numbers: 1,-1, 2,-2. The boundary condition is that the vertices lying opposite each other on the boundary (i.e., if the line passing through them goes through the center of Q) are labeled by opposite numbers (see Figure 9). Prove that there is an edge of the triangulation whose endpoints are labeled by opposite numbers, i.e., either an edge of type (1, —1) or an edge of type (2, -2). 7. The squares of a chess board (either a usual 8 by 8 board, or any other rectangular board of size m by ) are divided between a king and a rook in such a way that both pieces can move (according to the usual rules) just over its own squares (see Figure 10). Suppose there is no square on the extreme left-hand side of the chessboard from which the king can get to the extreme right of the board. Prove that then the rook can always find its way from the bottom row to the top row. Is the statement true if the king is replaced by a queen? And if both pieces are rooks? 8. (The “Hex” game.) The Hex game is played on a rhombus-shaped board made up of regular hexagons (see Figure 11). The board usually consists of 11 by 11 hexagonal fields, but for our purposes it may be of any size. A pair of opposite sides of the board belongs to the player who has the white pieces, rik[k{k | k[k{k]r kir{rf{kirirj{rjr r{k]r[k[r[r[k[k ki kiki ki] k[rpkir rir|[ef{r{r[el{kfr kir|kikjkikjrj{k ki} r[efrjk[k[rf{k kK{k]kir|rjk{k{r Ficure 104, SPERNER'S LEMMA 13 Ficure 11 and the sides are therefore called “white.” The other two sides of the board belong to the player who has black pieces, and are called “black.” The four corner fields belong to the two sides they lie in. Each player in turn places a piece of his or her color on any free field of the board. The goal of each player is to join his or her sides of the board with an unbroken chain of his or her pieces. For example, in Figure 11 the blacks have won. Prove that any hex game must end with a winner, i.e., there can be no draws. 9. A square Q is divided into smaller squares (faces) by lines parallel to its sides. The boundary of the square Q is oriented. That is, a direction is chosen on it such that if you move in that direction along the boundary, the internal points of the square Q always stay on your left (or on your right). For each of the other edges of the subdivision a direction is specified such that for each vertex inside Q there are exactly two edges coming to the vertex and two edges going from it (see Figure | 2). Prove that there is at least one oriented face in the subdivision (in Figure 12 there is exactly one such face, namely in the extreme lower right). 10. A polygon is a flat figure made up of a finite number of triangles (faces) in such a manner that (a) any two faces have no points in common or have a common vertex or a common side (edge); (b) any vertex can be reached from any other vertex by moving along the edges; (c) the figure has no holes inside. Thus, we assume that a polygon is defined by its triangulation. Jay} att J ya tt yoy aap yt Ficure 1214 4. SPERNER’S LEMMA The multiplicity of a vertex is the number of edges for which it is an end- point. Prove that the number of vertices of odd multiplicity in a triangulation of a polygon is even. 11. In Sperner’s lemma we are looking for the faces of a triangulation labeled by three different numbers, for example, 0, 1, and 2. Consider a tri- angulation such that (a) each of its faces is labeled by three different numbers 0, 1, and 2. Prove that then (b) all faces of the triangulation can be marked by two colors (black and white) so that no two adjacent faces (i.e., faces hav- ing a common edge) are marked by the same color. Prove that, conversely, (b) implies (a). 12. A triangle is triangulated in the following manner: each of its sides is divided into m equal parts (m is an arbitrary natural number), and the lines parallel to the sides of the triangle are drawn through the subdividing points (see Figure 13). Each vertex of the triangulation is colored either red or blue. Prove that the number of edges of the triangulation having endpoints of different colors is even. Does the statement hold for any triangulation of the triangle? If not, what is the class of triangulations for which it is valid? Ficure 135 Continuous Mappings, Homeomorphisms, and the Fixed Point Property As we have already said, a mapping is continuous if it takes points close to each other into points that are also close to each other. Now we have to make this definition more precise. Whether points are close to each other or not depends on the distance between them. A point y is said to be close (or sufficiently close) to the point x if the distance between them is less than a fixed number r > 0. The set of such points is said to form the r- neighborhood of the point x. In geometrical terms, an r-neighborhood on a plane is a disc without boundary of radius r with center at x. In three- dimensional space a neighborhood is a ball without its boundary surface. If X isa plane set (line, circle, square, etc.), and x is one of its points, then the r-neighborhood of the point x in the set X is, by definition, the set of all points in X whose distance from x is less than r. Therefore, such a neighborhood is the intersection of the set X and the disc of radius r (see Figure 14). Now let X and Y be two sets (on a plane) and f a mapping from X to Y. This is denoted as follows: f: X + Y. If x isa point of X and y = f(x), then y, as we already know, is said to be the image of the point x, while the point x is called the preimage of the point y. Ficure 14 1516 5. CONTINUOUS MAPPINGS, HOMEOMORPHISMS, FIXED POINT PROPERTY The mapping f: X — Y is said to be continuous at a point x if for any e >0 there exists a number 5 > 0 such that the image of each point in X from the d-neighborhood of the point x lies in the e-neighborhood of the point y. Thus, the continuity of a mapping at a point x means that for any ar- bitrary neighborhood of the image point y there is a neighborhood of the point x, corresponding to the first neighborhood and depending on it. Here the term “corresponding” means that all points of the second neighborhood are mapped into the first one. If a mapping f: X — Y is continuous at each point of the set X then it is said to be continuous on X or simply continuous. Example |. The projection of a square on its side AB is the mapping taking each point x of the square into the endpoint f(x) of the perpen- dicular dropped from x on AB (see Figure 15). For any e-neighborhood of the point f(x) the d-neighborhood of the point x of the same radius projects into the first neighborhood. Hence, for any ¢ > 0 there exists a corresponding 6 > 0, in this case equal to ¢. Therefore the projection map is continuous at each point of the square. Example 2. Let K, and K, be two concentric discs with a common center at the point O and radii r and R > r, respectively. Define a mapping f : K, + K, as follows. The point O goes into itself, while each radial segment OA of the disc K, is mapped linearly onto the radial segment OB of the circle K, going in the same direction (see Figure 16). Linearity of the map means that if a point C goes into the point D, then oc _ op OA OB If, under the same transformation, a point E goes into the point F, the similarity of the triangles OCE and ODF (see Figure 16) implies that the Ficure 155. CONTINUOUS MAPPINGS, HOMEOMORPHISMS, FIXED POINT PROPERTY 7 Ficure 16 length of the interval CE is increased by the factor of R/r. Therefore, a circle goes into a circle, and a disc of radius 5 goes into the disc of radius é€ =(R/r)d. Thus, for any e > 0 there is the corresponding 5 = (r/R)e, which means that the mapping is continuous. Example 3. Let K bea disc of radius R with the center at a point O. Consider the central projection of the disc K onto its boundary circle C. It is defined in the following way. Let x be a point of the disc other than its center. Then f(x) is defined as the point at which the ray issuing from the center O and passing through the point x intersects the circle C (see Figure 17). No mapping is defined for the point O in that way, but we shall assume that the center is projected, for example, into the topmost point 4 of the circle C. We now show that this projection is discontinuous at the center of the disc and continuous at any other point of the disc. Take an e-neighborhood of the point A = f(O) on the circle C, sufficiently small so that it does not contain the entire circle. One can, for example, take any Ficure 1718 5. CONTINUOUS MAPPINGS, HOMEOMORPHISMS, FIXED POINT PROPERTY e <2R. There is no corresponding 6 for this value of ¢. However small a 6-neighborhood of the point O is, the points of this neighborhood are mapped into points scattered all over the circle, and hence cannot possibly fit into the chosen e-neighborhood of the point O. Thus, the mapping is discontinuous at the center of the disc. Let x be a point on the disc other than its center (see Figure 17), f(x) its image, and the arc MN (without the endpoints) the e-neighborhood of the point f(x). Then the corresponding 6-neighborhood of the point x can be easily found in the following way: take the largest disc with its center at x lying in the circular sector subtended by the arc MN. The radius of this circle is 6. One can derive a formula expressing the dependence of 6 on @ (see Problem 14). We say f: X + Y isa one-to-one mapping if it takes different points of set X into different points of set Y. A mapping f: X + Y is said to bea surjection if each point of set Y is an image of some point (or points) of set X . Thus, under a surjection, set X is mapped on the entire set Y , not just part of it. For that reason a surjection is sometimes referred to as an onto mapping. For a one-to-one surjection f : X -+ Y we define the inverse mapping fi : Y +X associating to each point y in Y the point x in X that is mapped by f into y. A one-to-one correspondence f : X — Y is said to be a homeomorphism if it is continuous and the inverse mapping f— ' is also continuous. Sets X and Y are then said to be homeomorphic. A property of a set preserved under any homeomorphism is a topological property. Properties of this kind are studied by the science of topology. It follows from Example 2 that any two discs are homeomorphic (try to carry out the proof yourself!). One can verify that any disc is homeomorphic to a square, a triangle, or a polygon. The boundaries of these figures are also homeomorphic (i.e., a circle is homeomorphic to the boundary of a square or a triangle). Aset X is said to have a fixed point property if each continuous mapping f:X — X of this set into itself has a fixed point. We can easily see that this property is a topological one. That is, if a set X has it, then so does any set Y homeomorphic to X. To check this, let g : X — Y be any homeomorphism, and let f : Y + Y be any continuous mapping. Consider the mapping g = gt fg of the set X into itself. This mapping (which is called the composition of the mappings g, f, and g! ) is defined in the following way. First, the map g is applied to a point x in X. The result is the point y = g(x) in Y. Then f is applied to y, yielding the point S(y) = fa(x) in Y. Finally, after the application of gs one gets the point gt fg(x) in X . One can verify that the mapping g of the set X into itself is continuous. Therefore, it has a fixed point x, in X . Denote the image of this point under the homeomorphism g by yp, i.e. Yo = 8(%). We have 9(%) = 871 £8(%) = Xo.5. CONTINUOUS MAPPINGS, HOMEOMORPHISMS, FIXED POINT PROPERTY 19 Applying g to both sides of this equality, we get 88 '£8(%) = 8(%); or (5.1) £8(Xq) = 8(%)- Because g(X,) = Yo, formula (5.1) implies that f(y) = yy, So that yy isa fixed point. The importance of the statement we have just proved cannot be overesti- mated. Now we can always verify the fixed point property for just one repre- sentative of an entire class of homeomorphic sets. For example, in Chapter 8 this property will be proved for a square, and thereby it will be automatically verified for a disc, triangle, or any convex plane polygon whatsoever. Problems. 13. Let X and Y be two sets (on a plane), and let f bea mapping from X to Y. What can be said about this mapping if (a) for any 5 > 0 and any e > 0 the image of each point from the 6-neighborhood of some point x, lies in the e-neighborhood of the point Yo = S(%)s : ° (b) for any 5 >0 there exists ¢ > 0 such that the image of each point in the 5-neighborhood of the point x, lies in the e-neighborhood of the point Vo = S(X)? ° 14. Prove that the radial projection of a disc on its boundary circle is continuous at any point other than its center. Find the formula expressing the dependence of 6 on e. 15. A closed interval [a, b} is said to be nondegenerate if a # b. Prove that all nondegenerate closed intervals are homeomorphic. 16. Prove that the composition of two (or three) continuous mappings is a continuous mapping.6 Compaciness Let {x,,%2,-.. ,X,,---} bea sequence of points ona line, ona plane, or in a three-dimensional space. We shall denote such a sequence by {x,}. The individual element of the sequence numbered with n will be denoted x,. A sequence {x,} is said to converge to a point x, as n tends to infinity if for any € > 0 there exists an integer m such that for all n > m the distance between the points x, and x, is less than e. It is denoted x, -> x). In other words, the convergence x, —> X) means that for any e-neighborhood of the point x, all the points of the sequence after a certain one lie in this neighborhood. For example, the sequence of points x, = (1+ (-1)")/n on a line converges to zero since the odd members of the sequence are equal to zero, while the member numbered by any even number n = 2k is of the form X,, = 2/2k = 1/k. For the inequality 1/k 1/e whence n/2 > 1/e, or n > 2/e. Thus, for any integer n greater than m =[2/e]+1 (where the square brackets denote the greatest integer in 2/e) the point x, numbered by 1 belongs to the e-neighborhood of zero. On the other hand, the sequence 11 1 1 1 {1, 1-5, z lage ‘Iea1° 1-30} does not converge, since it includes two subsequences converging to different points. Indeed, its subsequence eee eee ae converges to zero, while the subsequence {1 -1/2,1-1/4,1-1/6,...,1- 1/2k,...} converges to 1. With the help of sequences the notion of compactness, one of the most important in topology, is defined. Asset X is said to be compact if any sequence {x,,X,,...,X,,---} of points of X contains a subsequence {,, Xr eee Maser } converging to some point x of the same set X. 22 6 COMPACTNESS, For example, any finite set is compact because any sequence of its points contains a stationary subsequence, all the points of which are equal, and such a sequence evidently converges. We shall now prove that both a closed inter- val and a square are compact. All convex polygons have the same property, as well as a disc, a circle, and a sphere. Let us start with two examples of noncompact sets. They are the real line R and its open interval (0, 1). Indeed, the sequence of natural num- bers {1,2,3,...,”,...} in R has no convergent subsequence, so that the real line R is not compact. On the other hand, while the sequence {1/2, 1/3, 1/4,... , 1/a, ...} of points of the interval (0, 1) converges to zero, its limit (zero) does not belong to this interval whereby the compactness condition is again violated. The proof of compactness of a closed interval is based on the Principle of Nested Closed Intervals. This principle (or an equivalent statement) is usually taken as one of the axioms of the real line. PRINCIPLE OF NESTED CLOSED INTERVALS. If a sequence of closed intervals {Ig. 1,5 +++ 51, +++} is such that each interval lies inside the preceding one, and their lengths tend to zero as n tends to infinity, then all the closed intervals have a unique common point. Let us prove that the closed interval J) = [a, b] is compact. Let {x,,%),...,%,)+-.} be any sequence of points lying in [a, 6). Divide the interval [a,b] in half. At least one of the halves contains in- finitely many (either distinct or coinciding) points of the sequence {x, }. Denote this half by J, , and choose a point x, of the sequence {x,} in it. Divide the closed interval /, also in half, and denote by J, the half that contains infinitely many points of the sequence {x,}. Proceeding in the same way, we get the sequence of the “nested” closed intervals {J, = [a, 5}, 1,,...,4,5.-.}. The length of the closed interval /, equals (b - a)/2*, anu the closed interval J, contains a point x, of the se- quence {x,}. The lengths of the closed intervals 1), evidently tend to zero. According to the Principle of Nested Closed Intervals there is a unique point xX, belonging to all intervals J, (k = 0,1, 2,...), and to the in- terval [a,b] in particular. Let us prove that the subsequence of points {Xn, Xp re Xa ..} selected from the sequence {x,} tends to x). Take an arbitrary e> Oo "Then there exists an integer m such that the length of the closed interval /,, (equal to (6 — a)/2”) is less than ¢. The closed interval I,, contains the point x, and therefore lies inside the e-neighborhood of this point. The same is true for all closed intervals J, with indices k > m. Thus, the e-neighborhood of the point x, contains all points of the subsequence {x,,} starting with the index 1,,. Therefore, the sequence {x,,} converges to the point x), which proves that the closed interval J, is compact. Now we prove the compactness of a square. Let Q, be a square whose sides are parallel to the coordinate axes Ox and Oy and equal to 1. Let {p,} be an arbitrary sequence of points in Q,. Cut Q, into four equal squares3. COMPACTNESS, 23 0 %y x Ficure 18 with straight lines parallel to its sides. Denote by Q, a smaller square (one of these four) that contains infinitely many (either distinct or coinciding) points of the sequence {p,}. Choose one of these points and denote it by Pa, Cut the square Q, again into four equal squares, and denote by Q, the square that contains infinitely many points of the sequence {p,}. Choose one of them and denote it by Pry Proceeding in the same way, we get the sequence of nested squares {Q,, Q,,.-. , Q,,--.} such that the side length of the square Q, equals 1/2, and Q, contains the point p,, of the se- quence {p,} (see Figure 18). The projections of these squares on the Ox axis forms a nested system of closed intervals {1j, 1, .-.[gs--- be the unique common point of these intervals. Now let {I , I], ...2; be the sequence of nested closed intervals on the Oy axis that are the projec- tions of the squares {Q,,Q,,... , Q,,--.} omit. Denote by y, the unique common point of these intervals. We leave it to the reader to prove that the subsequence {P,,} converges to a point of the square Q, with coordinates (Xq» Yo) - This concludes the proof that the square Q, is compact. Problem. 17. Prove that if a function f is continuous on the closed in- terval [a, 5], then it is bounded from both above and below. That is, there are numbers m and M such that m < f(x) < M for all points x of the closed interval.7 Proof of Brouwer’s Theorem for a Closed Interval, the Intermediate Value Theorem, and Applications Brouwer’s theorem for a closed interval has been formulated in Chapter 1, We shall now give its proof using the rigorous notion of continuity in the “e—6 language.” The basic idea of the proof is to divide the closed interval into subintervals and look for a subinterval whose endpoints are moved by our mapping in opposite directions. These moves must necessarily be small, otherwise the mapping would be discontinuous. In other words, the subinterval can be considered as an approximate fixed “point.” The real fixed point is then “found” explicitly using the Principle of Nested Closed Intervals. Let f be a continuous mapping of a closed interval [a, 5) into itself. If at least one of the endpoints of [a, b] stays where it was, the theorem is proved. So let us assume that the mapping shifts both endpoints, point a to the right, and point 6 to the left. Divide the interval 1, = [a, 6] in half by the midpoint c = (a+ 5)/2. If point c stays where it was, the theorem is Proved, otherwise it moves either right or left. In both cases one of the halves of the closed interval J, has the endpoints that are shifted by the mapping f inside the corresponding interval. Denote this half by 7, . Repeat the process by dividing the interval /, in half. Then either its midpoint is a fixed point, or there is a half J, of the interval 7, with endpoints that are shifted inside it. Proceeding in the same way, either we find a fixed point in a finite number of steps or obtain an infinite sequence {I,,/,,/,,... ,1,,---} of nested closed intervals such that the function f shifts their endpoints inside the respective intervals. Let x, be the unique common point of these intervals. Let us prove that S(X) = Xo. This is a direct consequence of the continuity condition. To be more precise, suppose that the points x) and Yo = S(Xp) do not coincide, and that yo lies to the right of x,. The mapping f is continuous at the point x). In the e—6 language this means that for any é-neighborhood of the point y, there is a d-neighborhood of the point x, such that images of all points of it belong to the e-neighborhood of the point y,. The choice 2526 7. BROUWER’S THEOREM, INTERMEDIATE VALUE THEOREM, APPLICATIONS @ 43 4 HTS YE Yo Hte 6 Ficure 19 of ¢>0 (the radius of the neighborhood) is up to us. Choose it sufficiently small for the e- and d-neighborhoods to be disjoint (see Figure 19). Because all the points of the chosen d-neighborhood are mapped into the e-neighborhood, the function shifts all of them to the right. For a sufficiently large n the entire closed interval 1, lies inside the 6-neighborhood of the point x,, and therefore both endpoints move to the right, contradicting the fact that the left endpoint of J, moves to the right while its right end moves to the left. The proof is completed. As we explained in the Preface, the problem of finding solutions for the equation f(x) =0 in the closed interval [a, b] can be reduced to the prob- lem of finding fixed points of the mapping F where F(x) = Af(x) +x. Consider this situation in more detail, and let us first prove the following theorem. INTERMEDIATE VALUE THEOREM. Let f be a continuous function defined in the closed interval [a, b] such that f(a) < f(b). If a number c satisfies the condition f(a) < ¢ < f(b), then there is a point x, in the interval for which f(x9) =c¢. In other words, a continuous function in a closed interval takes all inter- mediate values. First, we shall prove the theorem in a special case when c = 0, and, consequently, f(a) < 0 and f(b) > 0. The idea of the proof is, starting from the function f(x) on [a, 5], to construct a new continuous function F(x) =Af(x) +x defined in the same closed interval where the parameter 4 # 0 is chosen in such a way that the function F(x) maps the closed interval [a, 5] into itself. This choice ensures that the function F(x) has a fixed point x, in the closed interval [a,b]. Then Af(x9) + x) = X, and, therefore, f(x) = 0. The process of choosing 4 is shown in Figure 20. The function F(x) maps the closed interval [a, 5] into itself if its graph lies inside the square ABCD. Therefore, the coefficient 1 must be sufficiently small in absolute value. In other words, the graph of f(x) must be contracted. Moreover, 4 must be negative because otherwise any contraction takes the graph of F(x) out of the square. Therefore, the problem is to determine the precise value of 2. We will find it in the following way. The continuous function f(x) is bounded on [a, 6]. That is, there exist m and M such that m < f(x) < M for any point x of the interval (see Problem 17). In our case one evidently has m<0 and M>O. Since f(a) < 0, the continuity of the function f(x) implies that f(x) <0 for all x sufficiently close to a. Choose a point x, such that f(x) <0 for all a< x 0 for all x, a if a 0. Take the inequality 4 > (a—x,)/M f(x) and multiply both its sides by f(x). Then Af(x) > (a— x,)/M f(x) , and, using the oe —f(x) >—-M, we have F(x) = Af(xy+x>2 _a-Xx, = oa oe Spite) +x ane M)+x=a-x, +x. We have f(x) > 0, so that the choice of the point x, implies that x > x, , and hence F(x) > a—x, +x > a@. Consider now the point x for which S(x) <0. Then Af(x) > 0 and hence F(x) =Af(x)+x>x>2a. Let us show now that if a > x > then F(x) < b. Consider a point x for which f(x) >0. Then Af(x) <0, and therefore F(x) =Af(x)t+x (b — x,)/mf(x) and multiply both sides by f(x). Then Af(x) < ((b—x,)/m)f(x), and using the inequality — f(x) < —m, we have F(x) = Ase) +x < 22x) 4x = Spey ax b-x, = Cm (-m) yim) +x = b-x, +x.28 7, BROUWER'S THEOREM, INTERMEDIATE VALUE THEOREM, APPLICATIONS Because f(x) <0, the choice of the point x, implies that x < x,, hence F(x) f(b) and the number c satisfies the inequalities f(a) > c > f(b). We suggest the reader to try to carry out the proof in this case as well. Finally, note that the fixed point theorem is a consequence of the interme- diate value theorem (Problem 18). Problems. 18. Derive the fixed point theorem for a closed interval from the intermediate value theorem. 19. Give a proof of the intermediate value theorem independent of the fixed point theorem. 20. Let f be a continuous function on the closed interval [a, b] that takes each endpoint of the interval into itself, that is, f(a) =a and f(b) =b. Let g be any continuous function taking the closed interval [a, 5) into itself. Prove there is a point x, in this closed interval such that f(x) = g(x). Is the same statement valid for an arbitrary continuous function g on [a, b]? 21. Prove that a continuous mapping f of a closed interval A = [a,, a,] onto the (entire) closed interval B = [b, , b,] containing A has a fixed point. Is the statement true if the image f(A) of the closed interval A does not coincide with B , but only lies inside it? 22. Let f be a continuous mapping of the closed interval [a, b] into itself. Prove that the mapping g = f 2 (the composition of the mapping f with itself) has at least two fixed points. Is the statement true if f maps the closed interval [a, 6] into itself, but not onto itself? 23. The function f defined on the closed interval [a, b] is said to be nondecreasing if x, < x, implies f(x,) < f(x), and nonincreasing if x, < x, implies f(x,) > f(x,). Prove that any nondecreasing (or nonincreasing) function that maps a closed interval into itself has a fixed point in the closed interval even if it is not necessarily continuous. 24. Prove that the cubic equation tS i GX +4,x° +4x+a,=0 where a, # 0, and the other coefficients are arbitrary real numbers, has at least one real root. For a, = 0 there is obviously the root x = 0. Can we tell if there is a positive or negative root, if a; #0? Let us now present some corollaries to the intermediate value theorem. Consider continuous mappings of a circle C into the real line R. The position of a point x on the circle is usually described by the angle a formed by the radius drawn from the center of C to x with the radius directed horizontally7. BROUWER’S THEOREM, INTERMEDIATE VALUE THEOREM, APPLICATIONS 29 Ficure 21 to the right of the center (see Figure 21). The angle a is said to be the angular coordinate of the point x. The counterclockwise direction is usually taken for the positive one. We shall assume for the time being that a varies between 0 and 2a (measured in radians). Two points x and x* lying on the circle are said to be antipodal if their angular coordinates differ by 2 (or if the chord connecting them is a diameter, which is the same). THE Borsuk-ULAM THEOREM FOR A CiRCLE. Let f: C + R be a contin- uous function defined on a circle. Then there is a pair of antipodal points x and x" such that f(x) = f(x"). Proor. Let C be the unit circle with its center at the origin, and a the angular coordinate of its point x (see Figure 21). Define a continuous func- tion g(y) on the closed interval J = [—1, 1] as follows. For a point y on I, let x be the point of intersection of the perpendicular raised from the point y on the horizontal axis with the upper semicircle, and let x* be the point antipodal to x. Let 8(y) = f(x) — f(x") = f(a) - fata). The values of the function g(y) at the endpoints of the closed interval J are 8(1) = f(0) - f(x) and g(-1) = f(m) — f(0). That is, g(1) = —g(—1). If g(1) =0, the theorem is proved. Otherwise the intermediate value theorem implies that there is a point y in the closed interval J such that g(y) =0. Now if x is the point of the circle corresponding to y then f(x) = f(x*). CoroLiary. At any given moment of time and on any great circle of the Earth (the equator, for example) there is a pair of antipodal points where the air temperature is the same. First PANCAKE THEOREM. If A and B are two bounded plane figures, then there is a line dividing each into two parts of equal area. 'K. Borsuk (1905-1982) was a Polish mathematician, and S. Ulam (1909-1984) was an American mathematician of Polish origin.30 7. BROUWER’S THEOREM, INTERMEDIATE VALUE THEOREM, APPLICATIONS Figure 22 This theorem assumes each of the two figures have an area (which is not necessarily the case, e.g. if a figure has a very complicated and intricate form). The significance of the theorem’s name is now clear. Two pancakes served on the same plate can be cut exactly in half with a single slash of a knife. For the proof let us put both figures inside a circle C (see Figure 22). (We can do this because both figures are bounded.) . For any point x on the circle C denote by D, the diameter of C passing through x. Let L, be perpendicular to D, , passing through the point on D,, located at the distance t from x (0 <¢U, be its vertices. Because the square Q is compact, one can select a subsequence of the sequence {x,,X2,... ,X,,--.} Converging to a point p, in Q. We can assume as before that it is the sequence {x,} itself. Then there are three more sequences that converge to the same point, namely {y,}, {Z,}, {u,}, because the length of the side of the square Q, tends to zero as n tends to infinity. Let us prove that py is a fixed point. Suppose points py and 4g = f(Po) are distinct, and consider two cases. First, let point q) lie strictly above Point py. In other words, g = 2/2 for py (see Figure 27a). (The angle 9 can be defined even for points p, that are not vertices of the subdivision). Draw a horizontal straight line L separating points p, and g,. Take an e- neighborhood of point g, satisfying only the condition that it has no common points with L. Since the mapping / is continuous at point py, there is a 5- neighborhood of py such that the images of all points from this neighborhood lie in our e-neighborhood of g,. We will choose this d-neighborhood to be small enough to have no common points with line L. Then the d-neighborhoods and the é-neighborhood not only have no points in common, but any point of the ¢-neighborhood of point q, lies above any Point of the 5-neighborhood of point p,. Hence, if p is any point (such as a vertex of the subdivision) in the 6-neighborhood of point py and g = f(p) is its image, then the displacement vector (p,q) is directed upwards with a possible tilt to the left or right. In other words, the angle g corresponding to point p satisfies the inequalities 0 < gy < 2. Therefore, such a point must be labeled by 1 or 2. On the other hand, for n sufficiently large, the vertices x,,¥,,Z,>U, of the square Q,, lie inside the chosen 6-neighborhood of point p, , so that none8 PROOF OF BROUWER’S THEOREM FOR A SQUARE 37 of them can be labeled by any number other than | or 2. This contradicts the fact that the vertices of Q, are labeled by at least three different numbers. We now consider the second case, where the displacement vector (py, qo)» looking upwards, tilts right so that one has 0 < g < 2/2 for point py. (See Figure 27b.) Draw two lines: L, horizontal and L, vertical in such a way that each of them separates the points p, and g,. Choose an e-neighborhood of point g) and a 6-neighborhood of point py as before, but subject to the condition that neither of them has a common point with L, or L,. If p is any point of the d-neighborhood of point py), and g = f(p), then the angle g corresponding to the point p satisfies the inequalities 0 < g < 2/2. Hence such a point may have only one label, namely 1. If one now considers, instead of point p, the vertices of the square Q, for a sufficiently large n, this again results in a contradiction. We have considered two cases corresponding to the first two rows of the table. The remaining six cases are left to the reader. Brouwer’s theorem for a square is proved.9 The Iteration Method Let f be a continuous mapping of the closed interval [a, 5] into itself. Consider the problem of solving the equation (9.1) x= f(x), approximately. In other words, we want to find an approximation to a fixed point of the mapping f. We can try to solve equation (9.1) as follows. Take for the “initial approximation” an arbitrary number x, inside the closed interval [a, 6] and substitute it into the right hand side of the equation. The value x, = f(x,) is taken for the second approximation. In general, for each successive approximation x, the next one, x,,,, is found by the formula x,,, = /(x,)- If a sequence of numbers {x,} (called the iteration sequence) has a limit Xo, then xq is an (exact) root of equation (9.1). Indeed, if n tends to infinity, the left side of the equality x, 41 = S(%,) tends to xq, while the Tight side tends to f(xo), whence x) = f(x). In practice, any process of finding iterations x, is terminated after a finite number of steps, namely, when one has x, * X,,, up to the chosen accuracy. Since x,,, = (x,), this means that, up to the chosen accuracy, x, ~ f(x,)» ie., x, is an approximation to the desired root. Such a scheme of finding an approximate solution of equation (9.1) is called the iteration method or the method of successive approximations. The iteration method has a convenient geometric interpretation. As we know (see Chapter 1) finding a root of equation (9.1) is equivalent to finding the abscissa of the common point of the graph of the function y = f(x) and the straight line y = x. Figure 28 on p. 40 shows both graphs together with the so-called iteration path. This path is constructed as follows. Draw the vertical line through point x, on the Ox axis up to point A, where it intersects the graph of y = f(x). The coordinates of A, are (x,,X) = (x,, f(x,)). Next draw the horizontal line through A, up to its intersection with the straight line y = x at point B,. The coordinates of this point are (Xy, Xq) = (f(x,), f(x,)). Again draw the vertical line through B, until it meets the graph of y = f(x) at point A, with the coordinates (x,, x;) = (x2, f(%)), ete. 3940 9. THE ITERATION METHOD Ficure 28 Figure 28 shows very clearly how the iteration process converges. The points A,, A,, A;,... approach the desired intersection point, and their abscissas x,, x,,X3;,-.. tend to the root of the equation. We can see easily that an iteration scheme does not necessarily converge in all cases. Consider, for example, the equation x = f(x), where f(x) = 1-x. Taking x, = 1 for the initial approximation, one has 3 1 BM EXS EMR SG so that the iteration sequence does not converge. The process is said to “go into a loop.” The reader will easily verify that in this case the iteration path consists of the four sides of the square and no longer resembles a “staircase,” or a “spiral* as in Figure 28. We now formulate the sufficient conditions for the convergence of an it- eration scheme. Let us introduce the following notation. A mapping f of the closed interval [a, b] into itself is said to be a contraction mapping (or simply a contraction) if it decreases the distance between any two points of this closed interval at least by a factor of M, where M > 1. Since the distance between the points x, and x, equals |x, — x,|, the condition can be formulated as follows. There is a number a, such that 0 2, and hence x, — X,. As we already know, this implies that x, is the root of the equation (9.1). It is the only root of this equation. Indeed, if f(xy) = Xo and f(y) = yo, then the contraction condition implies [xo — Yol = [£(%o) — FOp)1 < al% — Yol- Because a < 1, we have |x, — Yo| = 0 which means that the points x, and Yq coincide. Thus, if f is a contraction mapping of the closed interval [a, 5] into itself, then the iteration sequence converges to the unique root of the equation (9.1). An error estimate for the approximate value x, of the root can be obtained similarly to the derivation of inequality (9.3): [xq — Xo] Sa" lx, — X9- That is, the error decreases with the increase of n at least in a geometric progression with the ratio a. For more details on the iteration method the reader can consult the book [4] and the article [2]. Note that the condition for the convergence of an iteration process yields the following fixed point theorem: any contraction mapping of the real line into itself has a unique fixed point (see Problem 32). Problems, 29. Construct an iteration sequence (and draw the correspond- ing iteration paths) for the function f(x) = 1/x in the closed interval [1/2, 2] taking for the initial approximation x, = 1, and then x, = 2. 30. Suppose that the function f(x) is increasing and continuous in the closed interval [a, b], and that f(a) <0, f(b) > 0. Construct the itera- tion paths taking for the initial approximation x, = a, and then x, = 6.a2 9. THE ITERATION METHOD Prove that both iteration processes converge (although f is not necessarily a contraction). Do the roots obtained in both processes coincide? 31. Prove a more specified version of the intermediate value theorem. Let f(x) be a continuous function defined on the closed interval [a, 5]. Let A be the least and B the greatest value of f(x) in this closed interval. This means that f(x) >A for all x in the interval, and that there is a point x, in it such that f(x,) = A. Similarly, f(x) < B for all x and there is a point x, such that f(x,)=B. Let A4 Y. Because a square has the fixed point property, and a closed interval is a retract of a square, the theorem on retraction provides a new proof of the fixed point theorem for a closed interval. The same argument shows that all other retracts of a square (circle, triangle, or “tailed” square) also have the fixed point property (see Figure 29). FicureE 29 a44 10. RETRACTION One can easily see that a circle does not have the fixed point property. A rotation by the angle equal, say, to 2/2 displaces every point of the circle. On the other hand, a disc has the fixed point property, whereby we obtain an important corollary: there is no retraction of a disc onto its boundary circle (see Example 3). Thus, the statement that no disc can be retracted onto its boundary is a consequence of the fixed point theorem. Conversely, the fixed point theorem is an easy consequence of the theorem that there is no retraction of a disc onto its boundary. We give the proof by reductio ad absurdum. Suppose that there is a contin- uous mapping g: K — K of the disc K into itself that has no fixed points. That is, let_g(p) # p for all points p in K. For each p consider the ray connecting p with g(p) and directed from g(p) to p. Let f(p) be the point where the ray intersects the boundary circle C (see Figure 30). Then, associating to each point p the corresponding point f(p), one obtains the mapping f : K — C such that all points of the boundary circle C are its fixed points. This is a contradiction provided we can show that f is a continuous mapping. Take any e > 0 and consider the e-neighborhood of the point f(p) on the circle C (this neighborhood is evidently an arc). Since the mapping g:K — K is continuous at point p, for any a > 0 there is 6 > 0 such that each point from the d-neighborhood of point p is mapped into the a-neighborhood of point g(p). Take @ so small that any ray issuing from any point of the a-neighborhood and passing through any point of the d- neighborhood (which depends on a) has a common point with the circle C inside the e-neighborhood of the point f(p) (see Figure 30). This implies the continuity of the mapping f. Thus, we have deduced the fixed point theorem from the theorem stating that there is no retraction of a disc onto its boundary. To complete the proof we must provide an independent proof of the latter. However, we cannot dwell on this question here. FicureE 30