Class notes Sept. 25 Some examples on limits of sequences Exercise 2.3.3. (Squeeze Theorem): lim xn lim zn l , then lim yn l as well. n n If xn yn zn , for all n N , and if n Proof: Let 0 . Since the sequences {xn } and {zn } converge, there exits n0 N such that for all n n0 , | xn l | , and | zn l | . 2 2 Now, by Theorem 2.3.3, lim( zn xn ) 0 , and moreover, zn xn 0 for all n N . n Therefore, for 0 there exists n1 N such that for all n n1 , zn xn 2 (we don’t need in absolute value since the difference is positive). Let’s then take n max{n0 , n1} . We have | yn l || yn xn | | xn l | zn xn | xn l | 2 2 , proving that lim yn l . n Exercise 2.3.4. (Uniqueness of the limit of a sequence) If a sequence an converges, then its limit is unique. Proof: Assume lim an l1 and lim an l2 , and suppose l1 l2 . Then either l1 l2 , or n n l1 l2 . The two cases are similar so we will only analyze one of them. Suppose l1 l2 . d Let d l2 l1 0 . Since lim an l1 and lim an l2 , for there must exist n0 N n n 4 such that | an l1 | , and | an l2 | for all n n0 . d We have d l2 l1 | l2 an | | an l1 | 2 , impossible! Therefore the limit has to be 2 unique. Exercise 2.3.11. Césaro Means: If ( xn ) is a convergent sequence, then the sequence x x ... xn given by the averages yn 1 2 converges to the same limit. n Proof: Denote lim xn l . Here is the idea of the proof: since the sequence ( xn ) n converges, for a given 0 we will find a rank, say m N such that for n m, | xn l | 1 (1), where 1 will be defined shortly. We need to look at n | yn l || x j 1 n j (x j 1 j l) l || | . We will split the sum in the numerator into the sum up n n to m 1 and the rest of the terms. We have: m 1 n | yn l || (x j l) j 1 n | | x j l | j 1 n n | x j m l | (2) n n | x j l | (n m 1)1 1 (3). n n Since ( xn ) converges, it must be bounded (see Theorem 2.3.2), therefore there exists Now, according to (1), and the triangle inequality, j m j M 0 , such that | xn | M for all n N . Moreover, it follows that the limit is bounded by M as well, so | l | M . Again, using the triangle inequality, m 1 | x j l | m 1 (| x | | l |) 2(m 1) M (4). n n n 2(m 1) M 1 can be So in order to prove that ( yn ) converges we need to show that n made arbitrarily small. But if we evaluate what we have in the above expression, M is a fixed number, m will be again a fix number once we give ourselves 0 , but the j 1 j 1 j denominator n can get arbitrarily large. Therefore, if we define 1 n0 N , such that for all n n0 , 2 , we can find 2(m 1) M 1 (5). Putting together (3), (4) and (5), n 2 , for all n n0 , proving that lim yn l . n 2 2 Now let us look at the second part of the exercise: it basically says that if the Césaro means of a sequence converge, the sequence must not necessarily converge. We will show this by constructing a counterexample. Let xn ( 1) n . We have: the inequality in (2) becomes: | yn l | 1 x1 x2 ... xn , if n is odd yn n n 0, if n is even Then lim yn 0 , but the sequence ( xn ) does not converge. n 1 . 4 xn We can try to calculate a few terms to get a “feel” of the sequence. We have 1 3 x2 1, x3 , x4 . It looks like the terms are getting smaller and smaller. We see that 3 11 x2 x1 . We will conjecture that the sequence is decreasing, but before we prove it we will show that it is bounded. We prove that 0 xn 3, for all n 1 . This is obviously true for n 1 . Let us assume Exercise 2.4.2. We are given the sequence defined by x1 3, xn 1 that the inequality holds true for some n, that is, we have 0 xn 3 . Then 1 1 1 xn 1 , and therefore 0 xn 1 3 . So we proved that for 4 4 xn 1 0 xn 3, for all n 1 . 1 4 xn 4, so Next we show that the sequence is decreasing, xn1 xn , for all n 1 . We already checked the statement for n 1 . Let us assume that it holds true for some n, that is, we have xn1 xn . 1 We have xn 2 . Since we proved that the sequence is bounded above by 3, we 4 xn 1 have and by the induction assumption 4 xn 1 0 , 1 1 xn 1 xn , therefore xn 1 , so xn 2 xn1 . It then follows that the 4 xn 1 4 xn sequence is decreasing. Now, by the Monotone Convergence Theorem (Theorem 2.4.2), the sequence converges, so there must exist l lim xn . Since the limit exists, we may take the limit in the n 1 . Solving for l in the previous equation we 4l find two solutions, l1 2 3, and l2 2 3 . Now, both of them cannot be the limit of recurrence relation, which gives us: l our sequence, and since we proved that 0 xn 3, for all n 1 , l1 2 3 cannot be the limit of the sequence. Therefore lim xn 2 3 . n Exercise 2.5.1 (Theorem 2.5.2) Any subsequence of a convergent sequence converges to the same limit as the original sequence. Proof: Let ( xn ) be a convergent sequence, and let x lim xn . Let ( xnk ) be a subsequence of ( xn ) . For any 0 there exists a rank N 0 such that for any n N0 :| xn x | . Now, since ( xnk ) is a subsequence the indices nk are positive integers, so there must exist an index l 1 such that nl N 0 (more precisely, define l min{k : nk N0 } , that is, nl is the first index greater or equal to N 0 ). Then, for all k l , | xnk x | , and therefore the subsequence ( xnk ) has the same limit x. Below we will provide a more detailed proof of the Bolzano-Weierstarss Theorem. Theorem 2.5.5 (in Abbott) Every bounded sequence contains a convergent subsequence. Proof: Let ( xn ) be a bounded sequence. Therefore there exists M 0 :| xn | M for all n 1, or M xn M , for all n 1 . Consider the two subintervals [ M , 0] and [0, M ] . At least on of them must contain infinitely many terms of the sequence ( xn ) . Call that interval I1 [a1 , b1 ] (this is just a notation; of course we either have a1 M , b1 0 or a1 0, b1 M ). Now we define the first term in the subsequence to be the first in the original sequence that belongs to the interval I1 , that is, let n1 min{n : xn I1} . So we have xn1 I1 . Repeat the procedure for the interval I1 : consider the closed subintervals a b a b [a1 , 1 1 ],[ 1 1 , b1 ] . At least one of the subintervals contains infinitely many terms of 2 2 the sequence ( xn ) . Select the subinterval that does, and denote it I 2 [a2 , b2 ] . Define the second term in the subsequence xn2 I 2 , where n2 min{n n1 : xn I 2 } . We obtain in this way a recurrent definition of the intervals I k [ak , bk ] satisfying 1 I k I k 1 , and also bk ak (bk 1 ak 1 ) . Since b1 a1 M , it follows (by a simple 2 M induction argument) that bk ak k 1 . 2 We also obtain the subsequence ( xnk ), ak xnk bk , k 1 . We claim that this is a convergent subsequence. Let us first prove that the family of closed nested intervals I1 I 2 ... I k ... has a nonempty intersection, that is we prove there exists x :{x} I k . Indeed, we have the sequence ( ak ) of left end-points of the intervals, which is increasing and bounded (for every k, M ak M , and the sequence (bk ) of right end-points of the intervals, which is decreasing and bounded (for every k, M bk M ). Moreover, M bk ak k 1 0 . Therefore, by the Monotone Convergence Theorem (theorem k 2 2.4.2), both sequences have limits: there exist a lim ak , b lim bk , and a b . But since bk ak 0 , it follows that a b (details of this conclusion are shown in Lemma 1). So we found a point x a b in the intersection of the sets. Lemma 1 will also show that this is the only element of the intersection. We now prove that the subsequence ( xnk ) converges to x. Let 0 . There exists k 1 1 1 (indeed, just consider k 1 log 2 ). Then, for any k 1 2 l k :| xnl x | bk ak , therefore lim xnk x . such that bk ak QED Lemma 1. Let I k I k 1 , k 1 be the intervals constructed in Theorem 2.5.5. Then there exists x :{x} I k . Proof: We proved in Theorem 2.5.5 that there exist a lim ak b lim bk , where I k [ak , bk ] . If a b it follows that ak a b bk , for all k 1 , so [a, b] I k . Since 1 1 bk ak k 1 0 , there must exist some k 1 such that k 1 b a , but this 2 2 contradicts the fact that ak a b bk , for all k 1 . Therefore a b and the intersection I k contains a single point. QED