0% found this document useful (0 votes)
58 views133 pages

Metric Space

Uploaded by

Chippy Chippy
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
58 views133 pages

Metric Space

Uploaded by

Chippy Chippy
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
You are on page 1/ 133
1 Basic Concepts In many branches of mathematics, it is convenient to have available a notion of distance between elements of an abstract set. For example, the proofs of some of the theorems in real analysis or analytic function theory depend only on a few proper- ties of the distance between points and not on the fact that the points are in R or C. When these properties of distance are abstracted, they lead to the concept of a metric space. The notion of distance between points of an abstract set leads naturally to the discussion of convergence of sequences and Cauchy sequences in the set. Unlike the situation of real numbers, where each Cauchy sequence is convergent, there are metric spaces in which Cauchy sequences fail to converge. A metric space in which every Cauchy sequence converges is called a “complete” metric space. This property plays a vital role in analysis when one wishes to make an existence statement. Our objective in this chapter is to define a metric space and list a large number of examples to emphasise the usefulness and the unifying force of the concept. We also define complete metric spaces, give several examples and describe their elementary properties. In Section 1.5, we shall prove that every metric space can be ‘completed’ in an appropriate sense. 1.1. Inequalities The subject of inequalities has applications in every part of mathematics, and the study of metric spaces is no exception. In fact, the definition of a metric space involves an inequality which is a generalisation of the familiar triangle inequality, satisfied by the distance function in R. (|x — y| = |x — z| +|z— y| forall , y,zinR or.) In this section, we establish some inequalities that will be required for confirming that some of the examples we list are indeed metric spaces. Theses examples will be invoked repeatedly. Proposition 1.1.1. The function f(x) = ees x2 0, is monotonically increasing. B m4 1. Basic Concepts Proof. Let y > x = 0. Then 1 1 T4y STs Proof. Let x and y have the same sign. Without loss of generality, we may assume that x=0 and y=0, and so x+y xty x y l4|[xt+yl l4+xty l+xty l4+x4y 22,7 Hy bt +x 1+y l+fx° 1+fyl" Suppose x and y have different signs. We may assume that |x| >|y|. Then [x+y] Sle. It follows from Proposition 1.1.1 that Ietyl_ el ciel _ bl Tt|xty] 1+ix] 14+ fa] 1 +91 ‘This completes the proof. o ‘The next proposition is the well known arithmetic mean-geometric mean inequality, or AM-GM inequality, for short. Proposition 1.1.3. (AM-GM Inequality) If a> 0 and b> 0 and if0 0, is concave, we have In (aa + (1=d)b)=Alna + (1 =) Ind, ies Ind\b-* = In (Aa + (1 = d)D). As y = exp xisastrictly increasing function, it follows from the above inequality that ab! = hat (1-d)b. a Remark When x=0, y=0, p> land Vp + 1/q=1, we have LLL. Inequalities 25 1 1 steytyt ws ety ‘The inequality is obvious when either x or y is 0. If x # 0 # y, then it follows from Proposition 1.1.3 upon writing \ = I/p, a = x and bI-\ = y. Theorem 1.1.4. (Hélder’s Inequality) Let x; 20 and y,=0 for i= 1,2,...,, and suppose that p > 1 and q > 1 are such that I/p + fq = 1. Then In the special case when p = q =2, the above inequality reduces to ® nm \pn yh Yan<(E-) (E») as) ‘This is known as the Cauchy-Schwarz inequality. 4. To begin with, we Proof. We need consider only the case when 33x? £04 > assume that dsr i=t yy (4) 7 In this case, the inequality (1.2) reduces to the form Yayst (as) i To obtain (1.5), we put successively x= x; and y = yj for i=1,2,...,m in the inequality of the preceding Remark and then add up the inequalities so obtained. The general case can be reduced to the foregoing special case if we take in place of the numbers x, y; the numbers Gy" &)" for which the condition (1.4) is satisfied. It follows by what we have proved in the paragraph above that Ji Vays This is equivalent to (1.2). This completes the proof. a 26 1. Basic Concepts ‘Theorem 1.1.5. (Minkowsk’s Inequality) Let x;20 and y;=0 for i= 1,2... and suppose that p> 1. Then (S (at ») ue (= ) m (= 1) u as) Proof. If p = 1, the inequality (1.6) is self-evident. So, assume that p > 1. We write Soi: G+ 99? in the form owt yy oats tw ont ty. (7) Let q> Ibe such that 1/p + 1/q = 1. Apply Hélder’s inequality to the two sums on the right side of (1.7) and obtain y (Gq tyrs (S) . (S (+ wire) + (= # ) " (= (ity a) ‘0 (EN Ee)" JEen)" Dividing both sides of this inequality by (Of, (x; + y1)*)'"4, we obtain (1.6) in the case OH, (xi + yi)? £ 0. In the contrary case, (1.6) is self-evident. a a ‘Theorem 1.1.6. (Minkowski’s Inequality for Infinite Sums) Suppose that p= 1 and Tet {%nn21> (Ynlnz1 be sequences of nonnegative terms such that }7™ , x,? and Soe ya? are convergent. Then 7%, (Xn + yx)” is convergent. Moreover, = UP a Up foo \ MP (S40) =(S) +(Sx) : m= mt ms Proof. For any positive integer m, we have from Theorem 1.1.5, [Sem (E) (EY fat 1.2. Metric Spaces 7 Thus, {(S0"™, (in + yn)?)"?}, which is an increasing sequence of nonnegative real numbers, is bounded above by the sum (E4)"-E4)" Itfollowsthat 7%, (x + ya)” is convergent and that the desired inequality holds. Theorem 1.1.7. Let p > 1. For a=0 and b= 0, we have (a+ bP S2P(al + bP). Proof. If either a or b is 0, the result is obvious. So assume a > 0 and b > 0. Now the function that maps every positive number x into x? is convex when p > 1. So, (ary =e# 2 2 ie, (a+b)? S2P-\(a? + bP). a 1.2. Metric Spaces The notion of function, the concept of limit and the related concept of continuity play an important role in the study of mathematical analysis. The notion of limit can be formulated entirely in terms of distance. For example, a sequence {xn}, 1 of real numbers converges to xif and only if for all ¢ > 0 there exists a positive integer nny such that |%» — x] <¢ whenever n= m. A discerning reader will note that the above definition of convergence depends only on the properties of the distance Ja — | between pairs a, B of real numbers, and that the algebraic properties of real numbers have no bearing on it, except insofar as they determine properties of the distance such as, Ja - B] > 0 when a 4 B, Ja — 8] = 8 -Bl+/B-yI. There are many other sets of elements for which “distance between pairs of elements” can be defined, and doing so provides a general setting in which the notions of convergence and continuity can be studied. Such a setting is called a metric space. The approach through metric spaces illuminates many of the concepts of classical analysis and economises the intellectual effort involved in learning them. We begin with the definition of a metric space. and Ja 5 Definition 1.2.1. A nonempty set X with a map d: Xx X > R is called a metric space if the map d has the following properties: 28 1. Basic Concepts (MSD) diny)20 x yEX: (MS2) d(xsy) if and only if x = y; (MS3) d(x,y)=d(y,x) my EX; (MS4) d(xy)Sdiyz)+d(zy) -4YZEX. ‘The map d is called the metric on X or sometimes the distance function on X. The phrase “(X,d) isa metric space” means that d is a metric on the set X. Property (MS4) is often called the triangle inequality. ‘The four properties (MS1)-(MS4) are abstracted from the familiar properties of distance between points in physical space. It is customary to refer to elements of any metric space as poimts and (x,y) as the distance between the points x and y. ‘We shall often omit all mention of the metric d and write “the metric space X” instead of “the metric space (X,d)”. This abuse of language is unlikely to cause any ‘confusion. Different choices of metrics on the same set X give rise to different metric spaces. In such a situation, careful distinction between them must be maintained. ‘Suppose that (X.d) is a metric space and Y is a nonempty subset of X. The restriction dy of d to Y x Y will serve as a metric for Y, as it clearly satisfies the metric space axioms (MSI)-(MS4);s0 (V, dy) isa metric space. By abuse oflanguage, we shall often write (¥, d) instead of (Y, dy). This metric space is called a subspace of X or of (X, d) and the restriction dy is called the metric induced by don Y. Examples 1.2.2. (i) The function d:R x R— R* defined by d(x,y) = |x —y| isa metric on R, the set of real numbers. To prove that d is a metric on R, we need verify only (MS4), as the other axioms are obviously satisfied. For any x 742 € R, myth It is known as the usual or standard metric on R. Rays %q)ty ER, 1Si-+-5 Jn) in RY, define » we x,y) = (f (a= ») . iat (For 2 =2, d(xy) = ((x1 = y1)* + (02 — y2))¥? is the usual distance in the Car- tesian plane.) To verify that d is a metric on R", we need only check (MS4), ie. if 2= (22, we must show that d(x,y)Sd(x,z) +d(zy). For k=12.. (x-y) + (y-D= d(x, y) +d(y,2)- My = Xk Zep be = Zk = Yee ‘Then d(x, 2) = (S ‘) _ az, y) = ps x) _ and 1.2. Metric Spaces 29 . in d(x,y) = (= (ut a) . fi ‘We must show that oe ae Squaring both sides of (*), and using the equality (a+ by =a +2ab+P, we see that (*) is equivalent to Em-(E4) (Es) which is just the Cauchy-Schwarz inequality (see Theorem 1.1.4). This metric is known as the Euclidean metric on R”. When n > 1, R"" can be regarded as a subset of R" in the usual way. The metric induced on R"“! by the Euclidean metric of R" is the Euclidean metric of R'“!. (iti) Let X= R". For x = (x1, 425...) and ¥ = (is yh R", define . Mp p(x, y) = (5 ban vt) : where p= 1. Note that when p = 2 this agrees with the Euclidean metric, To verify that d, is a metric, we need only check that dj(x, y) = dy(x, 2) + dy(z, y) when 2 = (2y2%).-652)) ER" For k= 1,2,...4m set a= xe — 2h, be =m — yx. Then n Up n 4 dy(x2) = (Sim) > (zy) = (Siw ) ot = WP and an Up (x9) = (s tn= ni") : i ‘We need to show that (renr)"=(Ene) "(Zor)" However, this is just Minkowsk’s inequality (see Theorem 1.1.5). 30 1. Basic Concepts Here again, when n> 1, the metri corresponding metric of the same kind, duced by d, on the subset R'W! is the nea uw» p(x. y) = (= bx ~»') when x,y €R™. at (iv) Let X = R". For x = (x1,%2)-.-5%n) and y = (yts y2s---+ Yn) in R", define doo(3y) = max [xi = pil. ‘Then doy is a metric, Indeed, for z= (21, 225..+42n) €R", we have bi - yl S bai - il + li — may, bs — al + max lai — yi from which it follows that max, |s)—yi| = max fi — zi] + max |2i — vil ‘Remark. Examples (ii), (iii) and (iv) show that we can define more than one distance function, i.e, different metrics in the set of ordered n-tuples. (v) Let X be any nonempty set whatsoever and let asy={) EA? It may be easily verified that d isa metric on X. It is called the discrete metric on X. If Yis a nonempty subset of X, the metric induced on Y by dis the discrete metric on ¥. ‘Sequence spaces provide natural extensions of Examples (i), (iii) and (iv). (vi) The space of all bounded sequences. Let X be the set of all bounded sequences of numbers, ic., all infinite sequences such that If.x = {xhins and y= {yi}in1 belong to X, we introduce the distance d(x, y) = supilai— yl. Itis clear that dis a metric on X. Indeed, if 2 = {zj};21 is in X, then 1.2. Metric Spaces a1 bss — yl |i — 2d + Les 9d = sup,s: — 2] + supila — ye] = d(x,2) +d(z,y). ‘Therefore, (x, y) = sup;lxi — yi = d(x, 2) + d(z,y). (vii) The space (>. Let X be the set of all sequences x = {xi}; =1 such that (= a) coy PEL Ifx = {ai}ie1 and y= |yi}iz1 belong to X, we introduce the distance lp dix) = (Sh-»r) . Itis 2 consequence of the Minkowski inequality for infinite sums (Theorem 1.1.6) that d(x y) €R. Evidently, d(x, y)=0, d(x,y) =0 if and only if x=y, and d(x, y) = d(y,x). The triangle inequality follows from the Minkowski inequality for infinite sums. In the special case when p = 2, the space é, is called the space of ‘square summable sequences. (viii) The space of bounded functions. Let $ be any nonempty set and 8(S) denote the set of all real- or complex-valued functions on S, each of which is bounded, i.e., sup [f(a)| < oe. x If fand g belong to (5), there exist M > 0 and N > 0 such that sup |/()|=M and sup |g(x)|=N. ae xs It follows that sup [f(x) — g(x)| < 00. Indeed, x8 LF) = g(x) = [FC] + || = sup [F(2)| + sup [g(x)|, and so OS sup [f(x) — g(x) SM +N. Define (fg) = sup [fix)-gix)h fag € BS). 2 1. Basic Concepts Evidently, d(f, g) =0, d(f,g) =0 if and only if f(x) = g(x) for all x €$ and a(f,g) = d(g, f). It remains to verify the triangle inequality for 8(S). By the triangle inequality for R, we have UFC) = glx) = [fF = hd) + |) - g(x] =sup [Fd = hd] +sup [hGx) = g(x)| = d(f,h) + d(h,g) and so d(f.g)Sd(f,h) +d(h,g), for all f,g,h€ 9(S). The metric d is called the uniform metric (or supremum metric). (ix) The space of continuous functions. Let X be the set of all continuous functions defined on [a,b], an interval in R. For f,g € X, define A fog) = UP xcja, nlf) — g)]- ‘The measure of distance between the functions fand gis the largest vertical distance between their graphs (see Figure 1.1). Since the difference of two continuous functions is continuous, the composition of two continuous functions is continu- ous, and a continuous function defined on the closed and bounded interval [a,b] is bounded, it follows that d(f,g) € R forall f, ¢ € X. It maybe verified as in Example (viii) that d is a metric on X. The space X with metric d defined as above is denoted by Cla,bj. All that we have said is valid whether all complex-valued continuous functions are taken into consideration or only real-valued ones are. When it is necessary to specify which, we write Ce[a,b] or Cg[a,b]. Note that Cla, b] © ‘B(a, b] and the metric described here is the one induced by the metric in Example (viii) and is also called the uniform metric (or supremum metric). (x) The setofall continuous functionson [a,b] can also beequipped with the metric 5 aif.g)= [ Ife) = golds. Figure 1.1 1.2. Metric Spaces 33 g f [ > Figure 1.2 ‘The measure of distance between the functions f and g represents the area between their graphs, indicated by shading in Figure 1.2. If fyg € Clab], then If — g| € Cla, b], and the integral defining d(f, ¢) is finite. It may be easily verified that dis a metric on Cla, b]. We note that the continuity of the functions enters into the verification of the “only if” part of (MS2). % (xi) Let X= RU [oo] U { = co}. Define f: X + R by the rule if -co (aii) On X =RU {oc} Uf 00}, define (x,y) =|tan7! x —tan™! yl, x,y € X. (Note that tan~' (00) = 7/2, tan“ (— 00) = —77/2 and that the function tan~* is ‘one-to-one.) It may now be verified that X is a metric space with metric d defined as above. (xiii) The extended complex plane. Let X = CU {0}. We represent X as the unit ‘sphere in R°, S={(1x,5): xx ERM +349 = 1, and identify C with {(x1,x2,0): 11.22 € R}. The line joining the north pole (0.0,1) of Swith the point z = (x), x2) € C intersects the sphere S at (9,6) # (0,0, 1), say. We map C to S by sending z = (x1, x2) to (E,,{). The mapping is dearly one-to- ‘one and each point of § other than (0,0,1) is the image of some point in C. The correspondence is completed by letting the point oo correspond to (0, 0, 1). (See Figure 1.3.) at 1. Basic Concepts Froune 1.3 Analytically, the above representation is described by the formulae Corresponding to the point at infinity, we have the point (0,0, 1). Also, E+in We define the distance between the points of X by 2la zl “This is actually the chordal distance between those points on the sphere correspond- ing to the points z1,z2 € CU {oo}. Evidently, d(zi, 22) =0 and d(q,2) = 0 if and only if 2 = a. Also, d(2s, 22) = d(22, 2). For the triangle inequality, the following ‘elementary inequality will be needed: when 71,2, €C (2,2) = when 2) € C and z) = 0c. (1+ 2w)(1 42%) = 1 + 2w 4 20 4 |Z) + 2R(2w) + lal? }wl? 1+ 2felforl + [2P|o S14 [ef + [WP + [ef =(4F [20+ |wf). (as) From the identity (21 — 2)(1 + 252s) = (21 = 25)(1 + 2520) + (2s — 2)(1 + 221), 1.2. Metric Spaces 35, we have [a = zal(1 + las?) Sle = all + 25201 + [es — lll + 25211 Sha al [P70 4 beh? $+ les — aol + fast?) + fa? (as) using inequality (1.8). Dividing both sides of (1.9) by (+ lafy"d + aP)'70 + as?) ‘we get the triangle inequality when none among z, 23,23 is oo. Other cases are left to the reader. (xiv) Let d be a metric on a nonempty set X. We shall show that the function ¢ defined by e(xy) = min (1, d(x,y)}, where x,y € X is also a metric on X. Since d(x,y)20, so e(x, y)=min 1, d(x, y)} 20. 1fx=y, then d(x,y) =0,and hence, e(x,y) =min{1,d(x,y)} =0. On the other hand, if e(x,y) follows that d(x,y)=0, and hence that x=y. This proves (MS1) and (MS2). d(x,y)=d(y,x), it is immediate from the definition of ¢ that ¢(x,y) =e(y,x). Thus, (MS3) also holds. It remains to verify only the triangle inequality (MS4). Let xy,z€X. Observe that e(x,y)=1 for all x,y€X; therefore, e(x,y) Se(x,z) + e(z,y) if either e(x,z) o e(z, y) is 1. But if both are less than 1, then e(x,2) = d(x, 2), and e(2, y) = d(z,y). Accordingly, e(xsy) = min {1, dix, »)} S d(x, y) Sd(x4z) + dlz,y) = e(x,2) + (2,7). Proposition 1.2.3. Let (X, d) be a metric space. Define d':X x X + R by d(x y) 1+ d(x,y)" ‘Then di is a metric on X. Besides, d’(x, y) < 1 for all x,y € X. day) = Proof. We need prove only (MS4). Suppose x, y,2€ X. Then d(x,y) = 0,d(x, 2) = 0,d(z, y) =O and d(x, y) = d(x,z) + d(z, y) because dis a metric on X. Now, _ diez) dey) 162) +1) = aD 14 dap (x2) +d(z,y) = Tedd tdiey PY Theorem 112 1 T 1+—_ +aea+day 1 _dlsy) 14) 1+ dy)" S dey) Nv 36 1. Basic Concepts Examples 1.2.4, (i) Let X = R. For x,y € R, define be= yl 1+h; a metric on R, different from the usual metric described by (x,y) = |x — yl in Example 1.2.1(i). (ii) The space of all sequences of numbers. Let X be the space of all sequences of numbers. Let x = {xj}21 and y = [yi}j21 be elements of X. Define sos lx, y) Late d(x,y) ‘We prove that the axioms (MS1)—(MS4) are fulfilled by this distance. It is imme- diate that (x,y) =0, d(x, y) =0 if and only if x = y and that d(x, y) = d(y, x). For the triangle inequality, we have = d(x,z) +d(z,y). In R?, define d(x, y) = [xy — ya), where x = (x,,x) and y = (yj, 2). For x = (0,0) and y = (1,0), we have d(x,y) = O although x y. Therefore, the “only if” part of (MS2) does not hold and so, d is not a metric on R’. It may be noted, however, that all other properties of a metric, including the “if” part of (MS2) do hold. In such a situation, d is called a pseudometric. More precisely, we have the following defin- Definition 1.2.5. Let X be a nonempty set. A pseudometric on X is a mapping of X x X into R that satisfies the conditions: (PMSI) d(x.y)=0 xy EX: (PMS2) d(xy)=0 ifx=y; (PMS3) d(x,y)=d(y,x) xy EX; (PMS4) d(x,y)Sd(xy2)+d@y) — mYZEX. Another example of a pseudometric space is the following: Example 1.2.6. Let X denote the set of all Riemann integrable functions on [2,5]. For f.g € X, define 13. Sequences in Metric Spaces a7, 4 afi) = | UF (x) = gla. ily verified that d is a pseudomewic on X, but it is not a metric on X. soy={¢ ifx=a 1 ifa 0, there exists a natural number fp such that d(xq,x) < € whenever n= to. In this case, we also say that {x,] converges to x, and write it in symbols as x, — x. If there is no such x, we say that the sequence diverges. A sequence is said to be Remark 1. By comparing the above with the definition of convergence in R (or C), we find that 4, —> x if and only if limy so d(x%,,x) = 0, where d denotes the usual metric in R (or C). Remark 2. In case there are two or more metrics on the set X, then it is necessary to specify which metricis intended to be used in applying the definition of convergence. Remark 3. If x € X and x’ € X are such that xy —+ x and x, — x’, then x= x". Indeed, for ¢ > 0, there exist natural numbers m and mz such that d(2%q,x) < &/2 whenever n= mand d(x,x") where x= (x1,4)...530) and y= (yisJ2y--+5Ju) are in R" and p=. Let af) = (xf, AP... a8), k= 1,2,..., be a sequence of clements in R" that converges to x = (x12. 13. Sequences in Metric Spaces 39, . up Jim atx.) = Jim (S i? -») For every €>0, there exists an integer ko(e) such that [x — | 00 for each j = 1,...,m. Let € > 0 be given. For each j there exists positive integer kj such that © bP al< ap for k= Consequently, * ve d(x, x) = (5 be -»") 0, there exists an integer ky such that [xf = 5] 0 and each j = 1,...4", there exists an integer k; such that kMaxl whenever n = =|. jd ch ~(?) Bat in the metric space under consideration, d(s,,x) <1/2 if and only if 4{p,x) = 0 if and only if x, =x. Thus, xz — x in (X,d) if and only if all the x, ‘except possibly finitely many, are equal to x. 7 (@) Let X denote the space of all sequences of numbers with metric d defined by St taal dx) =) Sipe Laith-al where x= {5 )j21 and y= (y)jay are im X. Let [2,21 = {14Ije abner bea sequence in X which converges to, say, x € X. In other words, we yl a, =SB= > Oasn— oo. APs he? ‘This means that, for € > 0, there exists an integer nig(¢) such that Sa by?=sl <€ whenever n= no(¢), y FAL al and so 1_ bal Tt hm —xl 0 is arbitrary, it follows that Timy 2. a(" = x5 for each j, On the other hand, let limy oo x” Consider any ¢ > 0. There exists an integer m such that «for each j. 13. Sequences in Metric Spaces a As the first sum contains only finitely many terms and limy..s. xf" = xj for each j, there exists np(s) such that ys mi F< § whenever = n(6). Hence, d(x'"), x) <¢ whenever n= m(¢). Thus, convergence in the space of all sequences is coordinatewise covergence. Avi) Let X=E(p=1) and let d(wy)=(CEile—yl’)'”, where x= (heer and y= {yeh are in &,. Let {x Moor = heehee bea ‘sequence in £, that converges to, say, x € fp, ie, Ue Jim d(x, x) = a(S [2 — “r) ‘Then, for € > 0 there exists a positive integer p(¢) such that oo up (= 1? = at) ce fA for all n = no(e). Since € > 0 is arbitrary, it follows that lim. x{” = xx for each k. ‘The converse is, however, not true, as the following example shows. Let xf" = x + Bim where 0 ifkg¢n Pe { ifkan. Clearly, [x1 = x4] = Bin = 0 ifm > k. Consequently, lim,,_. x(”” = x; for each k. However, a(x) = (= bo — 9) no for all n. Thus, x!" x in the fp-metric. ‘The following does, however, hold. ‘Theorem 1.3.4. Let {x}, be a sequence in f, such that lim,_.. xf" = x, for cach k, where x= [x]421 is an clement of é,. Suppose also that for every € > 0 there exists an integer mo(e) such that (3.6 meh Up ’) 0, Then for m= mo(e/4), we have oo uP © (= wer) << forall n, (1.10) 4 kt and hence, w we (= wer) xl wl HY sat sar ( yore i) by Theorem 1.1.7 ame kom {the aR I [fal 42°(6)" by (1.10) and (1.11). By hypothesis, lim,.... 3(” = 4; for each k Therefore, for the finite sum above we kknow that there exists my(e) such that P< (2)? whenever n= Soy? = al < (6) ibe = no(e). mi ‘Together with the preceding inequality this implics that (a9) 42)() 13. Sequences in Metric Spaces B or atae,2) < (+29) whenever n= m(e)- But (1 4.2°)'/P <(2P 4 20)? = (2p) P< (227) '/? = (4P)'/P = 4 Therefore, (x'"!, x) <& whenever n= ng(e). Since the existence of such g() has been established for an arbitrary ¢ > 0, ithas been shown that lim,,_.. d(x”, x) =0. Examples 1.3.5. (i) Let X = C[a, b], the space of all continuous functions on [«, b]. It does not matier whether we consider only real-valued functions or also allow ‘complex-valued functions to be in C[a, b]. Define . aga) = | Yee - geal. A sequence {f,}y=1 in X converges to f € X ifand only if, . Jim (ff) = lim J Vala) = f(eo|ax =. Convergence in this metric does not imply pointwise convergence, as the following ‘example shows: Let f(x) = 1 for all x € [0, 1] and when 0SxS* fla) = ont when P< x=, n ‘The functions f, are illustrated in Fig. 1.4. Now, 1 1 [itso -sotae= [a= mde= 20 as noe. 0 0 n So dfn f) + Oas n— 00, Le. fy + f im the metric space. But fa(0) = 0 for all n while f(0) = 1, so that f. f pointwise on [0, 1] (1,0) Figure 1.4 a 1. Basic Concepts Indeed, the notion of pointwise convergence of sequences of functions on [a bj, ie, lim j, =f if and only if lim j,(x) =f(x) Vx € la, bl, ‘cannot be described by any metric on a nontrivial class of functions (Dugundji [8; p.273]). Gi) Let X = Cla, b], and d(f,g) = sup {{f(2) = g(a)l:x € [a, B]}. Let [fader be ‘a sequence of elements of Cla, b] that converges to, say, f € Cla, b]; i.e, Jim Uff) = lim sup {\fn() = f(0|:x € (a, b]} = 0. ‘This means that for ¢ > 0 there exists an integer r(z) such that sup [[fal) — f(|:x € [a,b]} < © whenever n= m(e), ite, the sequence {fj},21 converges uniformly to f on [a,b]. The converse is also true. Indeed, if f, — f uniformly on [a,b], then for any ¢ > 0 there exists an integer ‘mg(£) such that sup {[fulx) — f(x)[:x € [ab]} < © whenever n= (2), ie limps dUfmf) = 0. Therefore, convergence of a sequence of functions in la, b] in the uniform metric is the same as uniform convergence. (iii) Let X denote the space %(S) of all bounded functions defined on a nonempty set Sand df, g) = sup {If(x) = g(2)|2x € Sh where f, ¢ € X. As in (ii) above, a sequence {fn} n= 1 in X converges to fif and only if Su — f uniformly on S. Therefore, convergence of a sequence of functions in ‘#() in ‘the uniform metric is the same as uniform convergence. The reader may fill in the necessary details. 14, Cauchy Sequences ()-\:| +>) In real analysis (function theory), we have encountered Cauchy's principle of convergence. (Recall that a sequence {},<1 of numbers is said to be Cauchy, or to satisfy the Cauchy criterion, if and only if, for all ¢ > 0, there exists an integer m(e) such that |x, — xm] <€ whenever m= n(e) and n> m(e). The Cauchy principle states that a sequence in R or C is convergent if and only if it is Cauchy.) ‘The principle enables us to prove the convergence of a sequence without prior Knowledge of its limit. ‘The real sequence 14. Cauchy Sequences 45 is such that for m= the distance between the terms is given by 1 1 \(-)- (53) which tends to zero as m, n tend to infinity. In other words, the real sequence {Xn}ne 1» Where x, — 1 — 1/2", satisfies the Cauchy criterion and hence converges by_ Cauchy's principle of convergence. A similar situation arises with sequences of functions; in fact, it comes up more often than with real or complex sequences. An extension of the idea of Cauchy ‘sequences to metric spaces turns out to be useful. Chet to ee) Definition 1.4.1. Let dbe a metric ona set X. A sequence {xy}, =1 intheset Xis saidto be a Cauchy sequence if, for every > 0, there exists a natural number 1 such that, A[Xpp%m) < whenever n= 1m and m= tl. 1 Qn Om” Remark 1. A sequence {x,} in R or Cis a Cauchy sequence in the sense familiar from elementary analysis if and only if it is a Cauchy sequence according to Definition 1.4.1 in the sense of the usual metric on Ror C. Remark 2. It is cumbersome to keep referring to a “sequence in a set X with metric d”, ‘especially if it is understood which metric is intended and no symbol such as d has been introduced to denote it. We shall therefore adopt the standard phrase “sequence in a metric space X”. ‘Examples 1.4.2. (i) The sequence [xn}yz1» where xn = 1+ 1/2 +... +1/m does not satisfy Cauchy's criterion of convergence. Indeed, z So, it is not the case here that |, — x»| — 0 for large m and n. (ii) In C{0,1], the sequence fi fiy fy--- given by fulx) = atx > x€ [0,1] is Cauchy in the uniform metric. For m2 1 the function me om (mast mtx n+x (m#x\(n¥x)" being continuous on [0, 1], assumes its maximum at some point x € [0,1]. So, A fns fa) = sup {Ifn(x) = fa2)|:x € [0, 1} =a 1, (nF mtx) EH A 46 1. Basic Concepts for large m and n. Moreover, the sequence {f,},21 converges to some limit. For fix)=x mx Ea 1 Wats) = f00)| = | Argo wre ‘Therefore, {f6}y21 converges to the limit f where f(x) = x for all x € [0,1]. Proposition 14.3” A convergent sequence in metric space ita Cauchy sequence. > a kod Proof. Let {xn} be a sequence in a set X with metric d, and let x be an element of ‘X such that lim,—.. x, = x. Given any ¢ > 0, there exists some natural number d (mdm im such that d(x,, x) < £/2 whenever n= np. Consider any natural numbers n and ‘msuch that n= tig and m2 np. Then d(x,,x) < 6/2 and d( xp» x) < €/2. Therefore = J! (An. A(X Xn) = A(Xp,X) + d(XmyX) < €/2 + €/2 =E. a ‘Does the converse of Proposition 1.4.3 hold? If a sequence {xy},=1 in a metric space (X, d) fulfills the Cauchy condition of Definition 1.4.1, does it follow that the sequence converges? Examples 1.4.4. (i) Let X denote the set of all rational numbers with the usual metric, namely, d(x, y) = |x— y| for xy € X. It is well known that the sequence 14, 141, 1.414, converges to V2. It is therefore Cauchy. However, it does not converge to a point of X. So, a Cauchy sequence need not converge to a point of the space. (ii) Another example of a Cauchy sequence that does not converge to a point of the space is the following: Let X = C[0, 1] with metric d defined by : ayy =| Visr=sorlas, fg € ct0.0. , Let {falne2 be the sequence in C[0,1] defined by Clearly, {fulne2 is a sequence in C[0,1]. Moreover, it is Cauchy because Ans fe) = [ Val) — fl) 12 w/t =| sols + | Splex)dx = (1/2) 1/2-1/m 1p-1n ‘The measure df,» fa) is the shaded area in Fig. 1.5. 14. Cauchy Sequences 7 ‘Suppose now that there is a continuous function fsuch that d(fy, f) > 0. It will be shown that this leads to a contradiction. Since 1 pin 12 an =| Wreatac+ f ten —seolar+ | [= fOo|ax 0 12-1] ha and d(finf) — 0 as n — 00, it follows that 12 1 [Pveoiac=0 ana I, [t= fled|de = 0. Since f is continuous, we see that f(x) =0 for 0 Jn) in") are complete metric spaces. The metric space (X, d), where X denotes the set of rationals and d(x, y) = |x — y| for x, y € X, has been observed to be an incomplete metric space (see Example 1.4.4(i)). That the metric space (X, d) of rationalsis incomplete also follows on considering the sequence {y}y = 1, Where as 1. Basic Concepts wrt ty ed Xn = fattest as this is a Cauchy sequence but it converges to the irrational number e. In our next proposition, we need the following definition. Definition 1.4.6, Let {x;},21 be a given sequence in a metric space (X,d) and let Am}ye1 be a sequence of positive integers such that m, pzh wherex = (x1, a2-+.5an)and y = (yi, y25.++5 Ye) arein R", isa complete metricspace. Proof. Let {xpos x" = (x0, x$",...,x4), denote a Cauchy sequence in (Gd), ise dye, x) — 0 as mm’ — oo. Then, for a given ¢ > 0 there exists aan integer ng(¢) such that 14. Cauchy Sequences 49. . up (Si -x"") 0, there exists a positive integer mo(c) such that x Ue dy, x) = (= wr) no(e). This implies |x{" — x\"| < ¢ for all n,m no(e); ie., for each k the sequence {x\”"},,. is a Cauchy sequence of real numbers. So by Cauchy's principle of convergence, lity». a." = x,, say. Let xbe the sequence (x, %3y...)- It will be shown that x € X and limp. x"? = x. From (1.15) we have y Yl? = fr < oP (1s) a for any positive integer N, provided that m, m= m(¢). Letting m — oc in (1.16), we obtain w Vk - alse mi for any positive integer Nand all n> no(e). The sequence {ek xf = ab} 1 is a monotonically increasing sequence which is bounded above and therefore has a finite limit 377, [x\”” - xz)’, which is less than or equal to €”. Hence, ~ Ye (Ene -»r) Se for all n= no(e)- (1.17) C= Observe that by Minkowski’s inequality, Proposition 1.1.6, and consequently, x € X. Moreover, xi) — x in X in view of (1.17) above. a Proposition 1.4.11. Let X denote the space of all bounded sequences and let lx, y) = supl|se— yx] 21 =A) bbe the associated metric. Then (X,d) is a complete metric space. 14. Cauchy Sequences 51 Proof. As the argument is simpler than that of Proposition 1.4.10, the reader is invited to fillin the details. a Proposition 1.4.12. The space ‘8(S) of all real- or complex-valued functions fon S, each of which is bounded, with the uniform metric d(f,g) = sup{|f(x)— g(2)|:x € S}, is complete. Proof. Let {fn} be a Cauchy sequence in ‘(S). For each s€ S, we have [fu(s) — fur(s)| = (fin fin)» so that the sequence {f,(s)} in C is a Cauchy sequence and therefore convergent. Define f:$ — C by f(s) = lim, ful). We shall prove first that f € 8(S) and then prove that lim, d(f f) = 0. Since 1 > 0, therefore by the Cauchy property of {f,}, there exists some n such that fu» fm) < L whenever n= mo and m= m. In particular, d(fis fag) < 1, and hence [fuls) — fuy(s)| <1 for all s € S, whenever nZ ty. Since fn, € B(S), there exists some M > 0 such that |fy(s)|=M for all s€ S. Therefore, Un(s)| = [faCs) = oS) + Ufnu(S)|S1+M Vs € S whenever n2 Mm. It follows that [f(s)| = lim, [fu(s)] $1 + M for every s € S. Thus, f € B(S), as claimed. ‘Now consider any ¢ > 0. By the Cauchy property of {f,}, there exists some rig such that Un fn) <© whenever n= n and m= ‘Therefore, Uinls) — fn(s)] <€ Ws € S whenever n= mq and m= ty. It follows upon letting m — 00 that [fuls) —f(3)| Se whenever s € Sand n= 1m. ‘This says that limp... fn» f) = 0. o Proposition 1.4.13. Let X = Cla, b] and dif, g) = supt|f(x) - g()|:asx 0 there exists an integer no(e) such that m2 ro(e) implies d(fn» f) = sup{|fn(x) — fu(x)|: a=x=b} f(x), say, as m+ 00. We have thus defined a function f with domain [a]. It remains to show that f € C[a, ] and that lim... dp, f) = 0. 52 1. Basic Concepts Since for every x € [4 bl, Vint) ~ fal) < © provided that 1m, n= no(¢), it follows upon letting m — oo that [ilx) - fl se (1.18) for all n= no(e) and all x € [a,b]. ‘To see why fis continuous, consider any x € [a, b] and any 1 > 0. According to what has been noted in the preceding paragraph, there exists an integer m(n) such that, for every x € [a,b], we have [fa(x) = f(x)| 0 such that Vin) = fru %0) | <} for |x— x9] < 3. (1.20) Since UFC) — F0)| = [FC — fn 29] + Yin) — fa 0)] + Uf 0) — Fa) it follows from (1.19) and (1.20) that |f(x) — f(x»)| 0 and let 1p be the least integer greater than Le. If m, n= ng then 14. Cauchy Sequences 53 “Thus, the sequence is Cauchy. To see why it does not converge, suppose that it were to converge if possible to, say, p €N. Let m be any integer greater than 2p. Then ‘n= m implies that ies Porliol d L— =. tn Po" pom” p 2p 2p “This shows that the sequence cannot converge to pand therefore does not converge atall. (iii) Let X = C U {oo} and let d be given by | 2la— al pa Vitlalyt tbat vitlar ‘Then (X,d) is a complete metric space. Let {2)}n=1 be a Cauchy sequence in (X,d). If the sequence {2,},,=1 contains the point at infinity infinitely often, then it contains a convergent subsequence, namely ‘the subsequence cach of whose terms is oc. In this case, the sequence {zn} n21 ‘converges to 00 by Proposition 1.4.7. On the other hand, if the point at infinity appears only finitely many timesin the sequence, then we may assume without loss of ‘generality that the sequence consists of points of C only, as the deletion (or insertion) of finitely many terms docs not alter the convergence behaviour of a sequence. Case |. If the sequence {|Z,|},=1 is unbounded, then for every natural number k there exists a term 2 of the sequence such that |zu| > k, where these terms can be chosen so that m1 > my k= 1,2,.... Now, 2 (eq, 20) = 2 Vit bal? We thus have limj.o 2m = 00 in (%d). By Proposition 1.4.7, it follows that Timpce Zn = 0. when 21,2 €C , when z; € Cand 2 = co. 2 0 ako. VIF¢kR Gse Il, The sequence {|z|},21 is bounded, say by M > 0. Let ¢ > 0 be given. There ‘exists rig € N such that m, n= 1 implies 2124 = Zl Sec. [1+ lanl? 1+ lem? Since |z,| then limyoo d(%m Yu) = 05 but (ps Yn) = d(Yns2in) for every m, and, therefore, limy—so d(yn2%n) = 0, so that {ya} ~ {4a}. Transitivity: If {xa} ~ {yn} and {ye} ~ {2a}, then limy ox dn, Ya) = 0 and limy-.x: d(Ya, Zn) =0. We shall show that lim. d(%y, Z,) = 0. Since D0 SA ( Xp, Zn) SCX Yn) + Aor Zn) for all n, it follows that OS Lim Xp, Z) Slim d(% yp Yq) + Lim d(Yqs Zn) = 05 ile Him ssc d(Xm Zn) = 0. . ‘Thus, ~ is an equivalence relation and X splits into equivalence classes. Any two members of the same equivalence class are equivalent, and no member of an ‘equivalence class is equivalent to a member of any other equivalence class. Let X denote the set of all equivalence classes; the elements of X will be denoted by %7, etc. Observe that if a Cauchy sequence {xy} has a limit x € X, and if {yn} is 56 1. Basic Concepts equivalent to {x}. then limy.>c Yn = x. This follows immediately from the follow- ing inequality: AV) S AY Xn) + AC% 9 2)- Moreover, if {x,} and {y,} are two nonequivalent sequences, then lim, %»# Timy se Yor For, if lim, oc %n = x= lim,_.20 Yn, then the inequality 0 = (xy, Yu) = (Xp, x) + d(x, Yn) Jeads to liryc d(n yx) = 0, contradicting the fact that {%} and {ya} are two nonequivalent sequences. The constant sequence (x,x,...,*,...) is evidently Cau- chy and has limit x. Define a mapping f:X + X as follows: f(x) = % where % denotes the equiva- lence class each of whose members converges to x. Thus the constant sequence (y5.-.55...) is a representative of X. In view of the observations made above, the mapping fis one-to-one. We next define a metric p in X. For 5 € X, set (% 7) = lim d(sm yu)» where {xn} € X and {yu} € 7. Observe that [d(x Yn) = Als Yn)| SA Xas Xin) + ACs Yn)s where [%} €% and [yn] € 7 and so {d(x yn)}nz1 is a Cauchy sequence of real numbers, Hence, limy« d(%m Jn) exists, for R is a complete metric space. We first show that p is well defined. Indeed, if {x} ~ {xn} and {y,} ~ {yn}, then Jim dp, x,) =O and lim d(ymy,) = 0. From the relation A(x Yh) S xp Xn) + UC Yn) + ACS Yh) and the relation (ns Yn) S dm X4) + U(p Yo) + Ap Yad it follows that ‘Tim d(%p Yu) = lim d(2%, ¥,)- ‘We next show that p is a metric on X. te d(%:.Yn)=0 for all n, it follows that lim, d(x,.y.) 20. Thus, plx,j) =0. If x=7, then p(z,7) =lim,..d(%Xp Yn), where {x,} € X, {yal EF and {x,} ~ {y,}- But this means (by definition of ~) that lim, d(x» Yq) =0. ‘Therefore p(X, j) =0. Conversely, if p(% 7) =0, then limy x. d(%p Yn) =0 (by definition of p) and hence {xy} ~ {yn} so that = J. Since (sip. yu) = d( yur xu) for all n, we have p(x, 7) = p(¥,). Finally, 15. Completion of a Metric Space 57 plz, Jim dln, 25) S Jim dln Jn) + lim An 20) = PRI) + G.Ds where {x4} € % {¥4] € and {24} €Z ‘The metric p defined in X has the property that p(%J) = plf(a).fy)) = dy) for all x,y € X, ic. fis an isometric embedding of X into X. Next we prove that every % € X is the limit of a sequence in f.X). For this purpose, suppose {x} is a Cauchy sequence in the equivalence class %. For any k € N, there exists a positive integer ny such that d(x,,,,) <1/k for n= nj. Let j, be the equivalence class containing all Cauchy sequences converging to Xyy ie Je = f(%pq)- Then, (3 f(8n)) = pCR A) = fim d(x) =F ‘Thus, X= limp vo flim) Finally, we show that X is complete. Let {%"} be a Cauchy sequence in X. Since each ¥ is the limit ofa sequence in AX), we can find 7 € f(X) such that p(#", ")) < 1. Then the sequence {j”)} in X can be shown to be Cauchy by arguing as follows: * 0G 7) = pl, 2) + pH, 2) + pH”, H™) dg peen,zim yt 7 m “The right hand side can be madeas smalllas desired by choosing mand nlarge enough, for {X") }is Cauchy. Since y isin f(X), thereexists somey,, € X suchthat f(y,) = jy. ‘Thescquence {y,} in Xmust be Cauchy because {j(”)] isa Cauchy sequence in X and fis an isometry. Therefore, {yq} belongs to some equivalence class € X. We shall show that limp. (%”, %) = 0. To this end, we takeanye > 0 and observe that CE, 8) = pH, H) + pH"), 3) < ‘ + oF", %) and 06H, %) = p(fYm),)) = lim dm y,) Se for sufficiently large 1, since {yn} is a Cauchy sequence in X. This implies that Timy20 p(%"", %) = 0, thereby completing the proof that X is complete. Uniqueness: Let (X*.d*) and (X**,d**) be any two completions of (X.d). We shall show that (X*,d*) and (X**,d**) are isometric. Let x* € X* be arbitrary. By the definition of completion, there exists a sequence {hye itn X such that lin, oo %q = x". The sequence {2%} ,21 may be assumed to belong to X"*. Since X** is complete, {x,},=1 converges in X** to x*, say, ie., Timy-oo an = x**. Define @: X* X** by setting o(x*) = x"*. It is clear that the mapping ¢ is one-to-one and does not depend on the choice of the sequence {Xn}n21 Converging to x*. Moreover, by construction, (x) =x for x € X and a(x"), gy") = d*(eet,y*) for all x4, y* € X*. ‘This completes the proof. a

You might also like