Limits Roughly speaking, lim an = L means that an approaches L as n grows. Here is the n→∞ precise definition of limit. Definition 1 (Limit) (a) A sequence of real numbers is a rule which assigns to each natural number n a real number an . We’ll denote it a1 , a2 , a3 , · · · or an n∈IN . (b) Let L ∈ IR and an n∈IN be a sequence of real numbers. Then lim an = L if n→∞ ∀ ε > 0 ∃ N ∈ IN such that an − L < ε whenever n ≥ N The sequence is said to converge to L. (c) A sequence is said to diverge if it does not converge to any real number. Remark 2 Here is what that definition of “ lim an = L” says. Suppose you concentrate n→∞ on a tiny interval centred on L. The interval can be arbitrarily small. (That’s the ∀ ε > 0 in the definition — the interval is (L − ε, L + ε).) Then as along as n is big enough an is in the interval. (That’s the ∃ N ∈ IN in the definition — if n ≥ N , a is in the interval.) Example 3 In Example 2 of the notes “A Little Logic” we saw that the statement ∀ ε > 0 ∃ N ∈ IN such that 1 n < ε whenever n ≥ N is true. This statement implies that the statement ∀ ε > 0 ∃ N ∈ IN such that is true for an = Definition 1, so 1 n an − L < ε whenever n ≥ N and L = 0. Of course (1) is exactly the definition of lim an = L in n→∞ lim 1 n→∞ n c Joel Feldman. (1) 2014. All rights reserved. =0 September 23, 2014 Limits 1 Example 4 In this example, we consider lim (−1)n . The sequence an = (−1)n n→∞ n n∈IN is −1, 1, −1, 1, −1, 1, −1, · · ·. We would guess that lim (−1) does not exist. So we fix any n→∞ real number L and show that lim (−1)n cannot be L. To do so, let U be the statement n→∞ ∀ ε > 0 ∃ N ∈ IN such that (−1)n − L < ε whenever n ≥ N We wish to show that U is false. To do so, we split it up into bite sized pieces, working from right to left, just as we did in the notes “A Little Logic”. Precisely, we let (see (2) below) ◦ S(N, ε) be the statement “(−1)n − L < ε whenever n ≥ N ”, and ◦ T (ε) be the statement “∃ N ∈ IN such that S(N, ε)” or ∃ N ∈ IN such that (−1)n − L < ε whenever n ≥ N ◦ Then U is the statement “∀ ε > 0 T (ε)”. U z z }| { T (ε) }| { S(N,ε) z }| { n ∀ ε > 0 ∃ N ∈ IN such that (−1) − L < ε whenever n ≥ N (2) We analyze U using the same three steps as in Example 2 of “A Little Logic”. ◦ Step 1: We find all N ’s and ε’s for which S(N, ε) is true. Fix any ε > 0 and any N ∈ IN. The statement S(N, ε) is true if all values of (−1)n , with n ≥ N , lie in the interval (L − ε, L + ε). Regardless of what N is, the set of all values of (−1)n with n ≥ N is exactly {−1, 1}. So S(N, ε) is true if and only if both −1 and 1 are contained in the interval (L − ε, L + ε). That is, S(N, ε) is true if and only if L − ε < −1 < 1 < L + ε Now, regardless of what L is, the distance from L − ε to L + ε is 2ε. As the distance from −1 to 1 is 2. If ε < 1, the interval (L − ε, L + ε) is shorter than the distance between −1 and 1 and so cannot contain both. So if ε < 1, S(N, ε) is false for all N ∈ IN. ◦ Step 2: Because S(N, ε)is false for all N ∈ IN when ε < 1, T (ε) is false for all ε < 1. ◦ Step 3: We conclude that U is false since, as we have just seen, T (ε) is false for at least one ε > 0. For example T 12 is false. In conclusion, (−1)n has no limit as n → ∞. c Joel Feldman. 2014. All rights reserved. September 23, 2014 Limits 2 Example 5 In this example, we consider lim n. The sequence an = n n∈IN is 1, 2, 3, n→∞ 4, 5, · · ·. We would guess that lim n does not exist. So we fix any real number L and n→∞ show that lim n cannot be L. That is, we show that the statement U in n→∞ U z }| { ∀ ε > 0 ∃ N ∈ IN such that n − L < ε whenever n ≥ N | {z } (3) S(N,ε) {z | T (ε) } is false. ◦ Step 1: We find all N ’s and ε’s for which S(N, ε) is true. Fix any ε > 0 and any N ∈ IN. The statement S(N, ε) is true if all values of n, with n ≥ N , lie in the interval (L − ε, L + ε). But, for any N ∈ IN and any ε > 0 and any L ∈ IR there are certainly n’s with n ≥ N that obey n > L + ε. So S(N, ε) is false for all N ∈ IN and ε > 0. ◦ Step 2: Because S(N, ε) is false for all N ∈ IN when ε > 0, T (ε) is false for all ε > 0. ◦ Step 3: We conclude that U is false since, as we have just seen, T (ε) is false for at least one ε > 0. In conclusion, n has no limit as n → ∞. To this point it is logically possible that a sequence has more than one limit. The following theorem rules that out and also generalises Example 5. Definition 6 (Bounded) A sequence an A > 0 such that |an | ≤ A for all n ∈ IN. n∈IN is said to be bounded if there exists an Theorem 7 (a) A sequence has at most one limit. (b) If a sequence converges, it is bounded. Proof: (a) We’ll prove this by contradiction. Suppose that the sequence an n∈IN converges both to b and to c and that b 6= c. By definition, given any ε > 0 (we’ll pick a specific ε shortly), ∃ Nb ∈ IN such that ∃ Nc ∈ IN such that c Joel Feldman. 2014. All rights reserved. an − b < ε whenever n ≥ Nb an − c < ε whenever n ≥ Nc September 23, 2014 Limits 3 So if n is bigger than both Na and Nb , an is simultaneously within a distance ε of b and within a distance ε of c. Consequently |b − c| = |b − an + an − c| ≤ |b − an | + |an − c| < ε + ε = 2ε That is, the distance from b to c is less than 2ε. This is true for any ε > 0. In particular, it is true for ε = 31 |b − c|. So |b − c| < 2ε = 23 |b − c| =⇒ 1 3 |b − c| < 0 which is impossible. (b) Suppose that the sequence an n∈IN converges to L. By definition ∀ ε > 0 ∃ N ∈ IN such that an − L < ε whenever n ≥ N In particular, choosing ε = 1, there is an N ∈ IN such that n ≥ N =⇒ |an − L| < 1 =⇒ |an | = an − L + L ≤ |an − L| + |L| < 1 + |L| This tells us that all an ’s with index n greater than N are bounded by 1 + |L|. Without the restriction to n ≥ N , we still have |an | ≤ max |a1 | , |a2 | , · · · , |a|N−1 , 1 + |L| Evaluating limits by directly verifying Definition 1 is OK for very simple sequences but it gets ugly and hard √ very quickly. Try applying Definition 1 directly to show that, 1+1/n+2/n2 for example, limn→∞ 3+4/n+5/n2 = 31 . It is much more efficient to have a list of simple limits that we already know together with a toolbox that allows us to build complicated limits out of simple ones. As examples of known simple limits, it is trivial that, for any a ∈ IR, lim a = a n→∞ and we have already seen, in Example 3, that lim 1 n→∞ n =0 The following few theorems provides a toolbox and then we’ll get some more simple examples. c Joel Feldman. 2014. All rights reserved. September 23, 2014 Limits 4 Theorem 8 (The Arithmetic of Limits) Let a, b, c ∈ IR, Ñ ∈ IN, and an bn n∈IR , and cn n∈IR be sequences of real numbers. Assume that lim an = a n→∞ lim bn = b n→∞ n∈IR , lim cn = c n→∞ and that c and all of the cn ’s are nonzero. Then (a) lim an + bn = a + b n→∞ (b) (c) lim an bn = ab n→∞ lim 1 n→∞ cn = 1 c (d) If an ≤ bn for all n ≥ Ñ , then a ≤ b. Warning 9 The assumption that an < bn for all n does not imply that lim an < lim bn . A counterexample is provided by an = 0, bn = 1 , n n→∞ n→∞ a = 0, b = 0. Proof of Theorem 8: We’ll start by writing out our given data. Note that the ε and N in “∀ ε > 0 ∃ N ∈ IN such that S(N, ε)” are dummy variables, just as x is a R1 dummy variable in 0 x dx. You may replace ε and N by whatever symbols you like. The hypotheses of this theorem say ∀ εa > 0 ∃ Na ∈ IN such that ∀ εb > 0 ∃ Nb ∈ IN such that ∀ εc > 0 ∃ Nc ∈ IN such that an − a < εa whenever n ≥ Na an − a < εb whenever n ≥ Nb an − a < εc whenever n ≥ Nc (4) (5) (6) We also know that any convergent sequence is bounded. So there exist constants A, B > 0 such that |an | ≤ A and |bn | ≤ B ∀ n ∈ IN (a) We are to prove that ∀ ε > 0 ∃ N ∈ IN such that |an + bn − a − b| < ε whenever n ≥ N So pick any ε > 0. We must prove that there is an n ∈ IN such that |an + bn − a − b| < ε whenever n ≥ N Observe that |an + bn − a − b| = (an − a) + (bn − b) ≤ |an − a| + |bn − b| c Joel Feldman. 2014. All rights reserved. September 23, 2014 Limits 5 We want the right hand side to be smaller than ε. So we apply (4) and (5) with εa = εb = and conclude that ∃ Na ∈ IN such that |an − a| < ε 2 ε 2 ε 2 whenever n ≥ Na ∃ Nb ∈ IN such that |bn − b| < whenever n ≥ Nb Choose N = max Na , Nb . Then whenever n ≥ N we have both n ≥ Na and n ≥ Nb so that |an + bn − a − b| ≤ |an − a| + |bn − b| < 2ε + 2ε = ε (b) We are to prove that ∀ ε > 0 ∃ N ∈ IN such that |an bn − ab| < ε whenever n ≥ N So pick any ε > 0. We must prove that there is a N ∈ IN such that |an bn − ab| < ε whenever n ≥ N Observe that |an bn − ab| = (an − a + a)bn − ab = (an − a)bn + a(bn − b) ≤ |an − a| B + |a| |bn − b| ε We want the right hand side to be smaller than ε. So we apply (4) and (5) with εa = 2B ε (the “+1” in the denominator is there just to make sure that we are not and εb = 2|a|+1 dividing by 0) and conclude that ∃ Na ∈ IN such that |an − a| < ∃ Nb ∈ IN such that |bn − b| < ε whenever n ≥ Na 2B ε 2|a|+1 whenever n ≥ Nb Choose N = max Na , Nb . Then whenever n ≥ N we have both n ≥ Na and n ≥ Nb so that |an bn − ab| ≤ |an − a| B + |a| |bn − b| < < ε ε 2B B + |a| 2|a|+1 ε + 2ε = ε 2 (c) We are to prove that ∀ ε > 0 ∃ N ∈ IN such that c1n − 1c < ε whenever n ≥ N So pick any ε > 0. We must prove that there is a N ∈ IN such that c Joel Feldman. 1 cn − 1c < ε whenever n ≥ N 2014. All rights reserved. September 23, 2014 Limits 6 Observe that 1 − 1 = |cn − c| cn c |cn c| To get an upper bound on this, we need a lower bound on the denominator. This denominator, |cn c| varies with n, but, for large n, it is close to c2 . In particular, applying (6) |c| 5|c| with εc = |c| 2 , for example (we could just as well use 10 or 6 ), we conclude that ∃ Nc ∈ IN such that |cn − c| < |c| 2 whenever n ≥ Nc and then |c| |c| |cn − c| |cn − c| = =⇒ ≤2 2 2 |cn c| c2 n ≥ Nc =⇒ |cn | = |c + (cn − c)| ≥ |c| − |cn − c| > |c| − To get this smaller than ε, we apply (6) again, this time with εc = ∃ Ñc ∈ IN such that |cn − c| < c2 ε 2 c2 2 ε, conclude that whenever n ≥ Nc and choose N = max{Nc , Ñc }. Then whenever n ≥ N we have both n ≥ Nc and n ≥ Ñc so that c2 1 ε |c − c| 1 n − ≤2 < 2 22 = ε cn 2 c c c (d) We’ll prove this by contradiction. Assume that an ≤ bn for all n, but that the limit b < a. Then (4) and (5) with εa = εb = |a−b| imply that 2 ∃ Na ∈ IN such that |an − a| < ∃ Nb ∈ IN such that |bn − b| < |a−b| 2 |a−b| 2 whenever n ≥ Na whenever n ≥ Nb So if n ≥ max{Na , Nb }, we have >0 z }| { a n − bn = a n − a + a − b + b − bn > 0 | {z } | {z } >− |a−b| 2 >− |a−b| 2 That is an > bn , which is a contradiction when n ≥ Ñ . c Joel Feldman. 2014. All rights reserved. September 23, 2014 Limits 7 Definition 10 (Monotonic Sequences) A sequence an n∈IN of real numbers is said to be monotonic if either a1 ≥ a2 ≥ a3 ≥ · · · or a1 ≤ a2 ≤ a3 ≤ · · · . Theorem 11 (Monotonic Sequences) Any bounded monotonic sequence of real numbers converges. Proof: Let’s assume that the sequence an n∈IN of real numbers is bounded and obeys a1 ≤ a2 ≤ a3 ≤ · · · . The other case is similar. Then the set A = an n ∈ IN is a nonempty, bounded set of real numbers. So a = sup A exists by the least upper bound axiom. We’ll prove that lim an = a. n→∞ Let ε > 0. Since a is the least upper bound of A, there must be an element of A that is strictly bigger than a − ε. Otherwise a − ε is an upper bound for A. Suppose that element has index N . That is aN > a − ε. Then n ≥ N =⇒ a − ε < aN ≤ an ≤ a =⇒ |an − a| < ε Theorem 12 (Squeeze Theorem) Let an of real numbers that obey: n∈IN , bn n∈IN and cn n∈IN be sequences (i) There is an n0 ∈ IN such that an ≤ bn ≤ cn for all n ≥ n0 . (ii) lim an = lim cn = L. n→∞ n→∞ Then lim bn = L n→∞ Proof: On a problem set. Now for a collection of simple examples. (Well, at least some of them are simple.) Theorem 13 (Examples) 1 p n→∞ n (a) If p > 0, then lim (b) If p > 0, then lim n→∞ √ n =0 p=1 (c) If |x| < 1, then lim xn = 0. n→∞ α (d) For any α ∈ IR and b > 1, lim nbn = 0. n→∞ √ (e) lim n n = 1 n→∞ c Joel Feldman. 2014. All rights reserved. September 23, 2014 Limits 8 Proof: (a) Let ε > 0. Choose N to be any natural number obeying N > N p > 1ε and n ≥ N =⇒ n1p ≤ N1p < ε 1 1/p . ε Then (b) First consider the case that p ≥ 1. By Problem Set 2, question #2(b), 0≤ √ n 1 n (p p−1 ≤ − 1) The right hand side converges to 0. So, by the squeeze theorem, lim n→∞ √ n Next consider 0 < p < 1. Then lim n→∞ √ n 1 p = lim q n→∞ n 1 = 1 p lim n→∞ q n 1 p p − 1 = 0. =1 (c) Let ε > 0. Observe that n √ x − 0 = |x|n < ε ⇐⇒ |x| < n ε We have already seen, in part (b), that lim n→∞ √ such that n ε > |x| for all n ≥ N . √ n ε = 1. So, as |x| < 1, there is an N ∈ IN (d) The case α ≤ 0 is trivial, since then nα ≤ 1. So assume that α > 0. Write an = Then (1 + 1/n)α an+1 = an b Pick any 1 < b̃ < b and observe that α 1 + n1 ≤ b̃ ⇐⇒ 1 + ≤ b̃1/α ⇐⇒ n ≥ b̃1/α − 1 −1 So if N is any natural number bigger than b̃1/α − 1 , 1 n nα . bn −1 ′ 0 ≤ aN+1 ≤ aN bb ≤ aN .. . b′ 2 b 0 ≤ aN+ℓ ≤ aN+ℓ−1 bb ≤ aN b′ ℓ b ′ 0 ≤ aN+2 ≤ aN+1 bb .. .. . . ′ Since 0 < b′ b < 1, lim ℓ→∞ b′ ℓ b = 0 and so lim an = 0 by the squeeze theorem. n→∞ (e) On a problem set. c Joel Feldman. 2014. All rights reserved. September 23, 2014 Limits 9