Sequences and Series

11 Sequences and Series Consider the following sum: 1 1 1 1 1 + + + +···+ i +··· 2 4 8 16 2 The dots at the end indicate that the sum goes on forever. Does this make sense? Can we assign a numerical value to an infinite sum? While at first it may seem difficult or impossible, we have certainly done something similar when we talked about one quantity getting “closer and closer” to a fixed quantity. Here we could ask whether, as we add more and more terms, the sum gets closer and closer to some fixed value. That is, look at 1 2 3 4 7 8 15 16 1 2 1 1 = + 2 4 1 1 1 = + + 2 4 8 1 1 1 1 = + + + 2 4 8 16 = and so on, and ask whether these values have a limit. It seems pretty clear that they do, namely 1. In fact, as we will see, it’s not hard to show that 1 1 2i − 1 1 1 1 1 + + + +···+ i = =1− i i 2 4 8 16 2 2 2 253 254 Chapter 11 Sequences and Series and then 1 = 1 − 0 = 1. i→∞ 2i There is one place that you have long accepted this notion of infinite sum without really thinking of it as a sum: lim 1 − 0.33333̄ = 3 3 3 3 1 + + + +··· = , 10 100 1000 10000 3 for example, or 3.14159 . . . = 3 + 4 1 5 9 1 + + + + + · · · = π. 10 100 1000 10000 100000 Our first task, then, to investigate infinite sums, called series, is to investigate limits of sequences of numbers. That is, we officially call ∞ X 1 1 1 1 1 1 = + + + +···+ i +··· i 2 2 4 8 16 2 i=1 a series, while 2i − 1 1 3 7 15 , , , ,..., ,... 2 4 8 16 2i is a sequence, and ∞ X 2i − 1 1 = lim , i i i→∞ 2 2 i=1 that is, the value of a series is the limit of a particular sequence. 11.1 Sequen es While the idea of a sequence of numbers, a1 , a2 , a3 , . . . is straightforward, it is useful to think of a sequence as a function. We have up until now dealt with functions whose domains are the real numbers, or a subset of the real numbers, like f (x) = sin x. A sequence is a function with domain the natural numbers N = {1, 2, 3, . . .} or the non-negative integers, Z≥0 = {0, 1, 2, 3, . . .}. The range of the function is still allowed to be the real numbers; in symbols, we say that a sequence is a function f : N → R. Sequences are written in a few different ways, all equivalent; these all mean the same thing: a1 , a2 , a3 , . . . ∞ {an }n=1 ∞ {f (n)}n=1 As with functions on the real numbers, we will most often encounter sequences that can be expressed by a formula. We have already seen the sequence ai = f (i) = 1 − 1/2i , 11.1 Sequences 255 and others are easy to come by: i i+1 1 f (n) = n 2 f (n) = sin(nπ/6) f (i) = f (i) = (i − 1)(i + 2) 2i Frequently these formulas will make sense if thought of either as functions with domain R or N, though occasionally one will make sense only for integer values. Faced with a sequence we are interested in the limit lim f (i) = lim ai . i→∞ i→∞ We already understand lim f (x) x→∞ when x is a real valued variable; now we simply want to restrict the “input” values to be integers. No real difference is required in the definition of limit, except that we specify, perhaps implicitly, that the variable is an integer. Compare this definition to definition 4.10.4. DEFINITION 11.1.1 ∞ Suppose that {an }n=1 is a sequence. We say that lim an = L n→∞ if for every ǫ > 0 there is an N > 0 so that whenever n > N , |an − L| < ǫ. If lim an = L n→∞ we say that the sequence converges, otherwise it diverges. If f (i) defines a sequence, and f (x) makes sense, and lim f (x) = L, then it is clear x→∞ that lim f (i) = L as well, but it is important to note that the converse of this statement i→∞ is not true. For example, since lim (1/x) = 0, it is clear that also lim (1/i) = 0, that is, x→∞ the numbers i→∞ 1 1 1 1 1 1 , , , , , ,... 1 2 3 4 5 6 get closer and closer to 0. Consider this, however: Let f (n) = sin(nπ). This is the sequence sin(0π), sin(1π), sin(2π), sin(3π), . . . = 0, 0, 0, 0, . . . since sin(nπ) = 0 when n is an integer. Thus lim f (n) = 0. But lim f (x), when x is n→∞ x→∞ real, does not exist: as x gets bigger and bigger, the values sin(xπ) do not get closer and 256 Chapter 11 Sequences and Series closer to a single value, but take on all values between −1 and 1 over and over. In general, whenever you want to know lim f (n) you should first attempt to compute lim f (x), n→∞ x→∞ since if the latter exists it is also equal to the first limit. But if for some reason lim f (x) x→∞ does not exist, it may still be true that lim f (n) exists, but you’ll have to figure out n→∞ another way to compute it. It is occasionally useful to think of the graph of a sequence. Since the function is defined only for integer values, the graph is just a sequence of dots. In figure 11.1.1 we see the graphs of two sequences and the graphs of the corresponding real functions. 5 4 3 2 1 0 f (n) = 1/n • • • • 0 • • • • • 5 • 10 1 5 4 3 2 1 0 1 f (n) = sin(nπ) 0 • • • • • • • • • 1 2 3 4 5 6 7 8 −1 Figure 11.1.1 0 −1 .. .. .. .. .. .. ... ... ... .. ... ... ... .... ... .... ...... ......... .................... .................................................................................................... f (x) = 1/x 0 5 10 f (x) = sin(xπ) .. ....... ........ ....... ....... .. ... .. ... ... .... ... ... ... ... ... ... ... .... ... .... . . . .... .... . .... .... ... ... . ... ... . ... ... ... ... ... ... ... .. ... ... ... .. .. ... ... ... ... ... ... .. ... .... .. ... ... . ... ... . . . . . . . ... ... ... . . . ... ... . . . . . . ... ... . ... ... . . . . . . . . . . ... . ... .. ... ... ... ... ... .. ... .... ... ... ... .. ... .. ... .... ... .. ... ... ...... ...... ...... ..... Graphs of sequences and their corresponding real functions. Not surprisingly, the properties of limits of real functions translate into properties of sequences quite easily. Theorem 2.3.6 about limits becomes THEOREM 11.1.2 Suppose that lim an = L and lim bn = M and k is some constant. n→∞ n→∞ Then lim kan = k lim an = kL n→∞ n→∞ lim (an + bn ) = lim an + lim bn = L + M n→∞ n→∞ n→∞ lim (an − bn ) = lim an − lim bn = L − M n→∞ n→∞ n→∞ lim (an bn ) = lim an · lim bn = LM n→∞ n→∞ n→∞ an limn→∞ an L = = , if M is not 0 n→∞ bn limn→∞ bn M lim Likewise the Squeeze Theorem (4.3.1) becomes 11.1 Sequences 257 THEOREM 11.1.3 Suppose that an ≤ bn ≤ cn for all n > N , for some N . If lim an = n→∞ lim cn = L, then lim bn = L. n→∞ n→∞ And a final useful fact: THEOREM 11.1.4 lim |an | = 0 if and only if lim an = 0. n→∞ n→∞ This says simply that the size of an gets close to zero if and only if an gets close to zero. ∞ n EXAMPLE 11.1.5 Determine whether converges or diverges. If it conn + 1 n=0 verges, compute the limit. Since this makes sense for real numbers we consider  1 x = lim 1 − = 1 − 0 = 1. x→∞ x→∞ x + 1 x+1 lim Thus the sequence converges to 1. EXAMPLE 11.1.6 Determine whether  ln n n ∞ converges or diverges. If it con- n=1 verges, compute the limit. We compute 1/x ln x = lim = 0, x→∞ 1 x→∞ x lim using L’Hôpital’s Rule. Thus the sequence converges to 0. EXAMPLE 11.1.7 Determine whether {(−1)n }∞ n=0 converges or diverges. If it converges, compute the limit. This does not make sense for all real exponents, but the sequence is easy to understand: it is 1, −1, 1, −1, 1 . . . and clearly diverges. EXAMPLE 11.1.8 Determine whether {(−1/2)n }∞ n=0 converges or diverges. If it conn ∞ verges, compute the limit. We consider the sequence {|(−1/2)n |}∞ n=0 = {(1/2) }n=0 . Then  x 1 1 lim = lim x = 0, x→∞ 2 x→∞ 2 so by theorem 11.1.4 the sequence converges to 0. 258 Chapter 11 Sequences and Series √ EXAMPLE 11.1.9 Determine whether {(sin n)/ n}∞ n=1 converges or diverges. If it √ √ converges, compute the limit. Since | sin n| ≤ 1, 0 ≤ | sin n/ n| ≤ 1/ n and we can use √ √ theorem 11.1.3 with an = 0 and cn = 1/ n. Since lim an = lim cn = 0, lim sin n/ n = n→∞ n→∞ n→∞ 0 and the sequence converges to 0. EXAMPLE 11.1.10 A particularly common and useful sequence is {r n }∞ n=0 , for various values of r. Some are quite easy to understand: If r = 1 the sequence converges to 1 since every term is 1, and likewise if r = 0 the sequence converges to 0. If r = −1 this is the sequence of example 11.1.7 and diverges. If r > 1 or r < −1 the terms r n get large without limit, so the sequence diverges. If 0 < r < 1 then the sequence converges to 0. If −1 < r < 0 then |r n | = |r|n and 0 < |r| < 1, so the sequence {|r|n }∞ n=0 converges to n ∞ n 0, so also {r }n=0 converges to 0. converges. In summary, {r } converges precisely when −1 < r ≤ 1 in which case n 0 if −1 < r < 1 lim r n = n→∞ 1 if r = 1 Sometimes we will not be able to determine the limit of a sequence, but we still would like to know whether it converges. In some cases we can determine this even without being able to compute the limit. A sequence is called increasing or sometimes strictly increasing if ai < ai+1 for all i. It is called non-decreasing or sometimes (unfortunately) increasing if ai ≤ ai+1 for all i. Similarly a sequence is decreasing if ai > ai+1 for all i and non-increasing if ai ≥ ai+1 for all i. If a sequence has any of these properties it is called monotonic. EXAMPLE 11.1.11 The sequence  2i − 1 2i ∞ = i=1 1 3 7 15 , , , ,..., 2 4 8 16 is increasing, and  n+1 n ∞ i=1 = 2 3 4 5 , , , ,... 1 2 3 4 is decreasing. A sequence is bounded above if there is some number N such that an ≤ N for every n, and bounded below if there is some number N such that an ≥ N for every n. If a sequence is bounded above and bounded below it is bounded. If a sequence {an }∞ n=0 is increasing or non-decreasing it is bounded below (by a0 ), and if it is decreasing or nonincreasing it is bounded above (by a0 ). Finally, with all this new terminology we can state an important theorem. 11.1 THEOREM 11.1.12 259 Sequences If a sequence is bounded and monotonic then it converges. We will not prove this; the proof appears in many calculus books. It is not hard to believe: suppose that a sequence is increasing and bounded, so each term is larger than the one before, yet never larger than some fixed value N . The terms must then get closer and closer to some value between a0 and N . It need not be N , since N may be a “too-generous” upper bound; the limit will be the smallest number that is above all of the terms ai . EXAMPLE 11.1.13 All of the terms (2i − 1)/2i are less than 2, and the sequence is increasing. As we have seen, the limit of the sequence is 1: 1 is the smallest number that is bigger than all the terms in the sequence. Similarly, all of the terms (n + 1)/n are bigger than 1/2, and the limit is 1: 1 is the largest number that is smaller than the terms of the sequence. We don’t actually need to know that a sequence is monotonic to apply this theorem— it is enough to know that the sequence is “eventually” monotonic, that is, that at some point it becomes increasing or decreasing. For example, the sequence 10, 9, 8, 15, 3, 21, 4, 3/4, 7/8, 15/16, 31/32, . . . is not increasing, because among the first few terms it is not. But starting with the term 3/4 it is increasing, so the theorem tells us that the sequence 3/4, 7/8, 15/16, 31/32, . . . converges. Since convergence depends only on what happens as n gets large, adding a few terms at the beginning can’t turn a convergent sequence into a divergent one. EXAMPLE 11.1.14 Show that {n1/n } converges. We first show that this sequence is decreasing, that is, that n1/n > (n+1)1/(n+1) . Consider the real function f (x) = x1/x when x ≥ 1. We can compute the derivative, f ′ (x) = x1/x (1−ln x)/x2 , and note that when x ≥ 3 this is negative. Since the function has negative slope, n1/n > (n + 1)1/(n+1) when n ≥ 3. Since all terms of the sequence are positive, the sequence is decreasing and bounded when n ≥ 3, and so the sequence converges. (As it happens, we can compute the limit in this case, but we know it converges even without knowing the limit; see exercise 1.) EXAMPLE 11.1.15 Show that {n!/nn } converges. Again we show that the sequence is decreasing, and since each term is positive the sequence converges. We can’t take the derivative this time, as x! doesn’t make sense for x real. But we note that if an+1 /an < 1 then an+1 < an , which is what we want to know. So we look at an+1 /an : (n + 1)! nn (n + 1)! n+1 nn an+1 = = = n+1 n+1 an (n + 1) n! n! (n + 1) n+1  n n+1 n =  n n+1 n < 1. 260 Chapter 11 Sequences and Series (Again it is possible to compute the limit; see exercise 2.) Exercises 11.1. 1. Compute lim x1/x . ⇒ x→∞ n! 2. Use the squeeze theorem to show that lim n = 0. n→∞ n √ √ ∞ 3. Determine whether { n + 47 − n}n=0 converges or diverges. If it converges, compute the limit. ⇒  2 ∞ n +1 4. Determine whether converges or diverges. If it converges, compute the limit. (n + 1)2 n=0 ⇒ ∞  n + 47 converges or diverges. If it converges, compute the 5. Determine whether √ n2 + 3n n=1 limit. ⇒  n ∞ 2 6. Determine whether converges or diverges. ⇒ n! n=0 11.2 Series While much more can be said about sequences, we now turn to our principal interest, series. Recall that a series, roughly speaking, is the sum of a sequence: if {an }∞ n=0 is a sequence then the associated series is ∞ X i=0 an = a0 + a1 + a2 + · · · Associated with a series is a second sequence, called the sequence of partial sums {sn }∞ n=0 : n X sn = ai . i=0 So s0 = a0 , s1 = a0 + a1 , s2 = a0 + a1 + a2 , ... A series converges if the sequence of partial sums converges, and otherwise the series diverges. EXAMPLE 11.2.1 n If an = kx , ∞ X an is called a geometric series. A typical partial n=0 sum is sn = k + kx + kx2 + kx3 + · · · + kxn = k(1 + x + x2 + x3 + · · · + xn ). 11.2 Series 261 We note that sn (1 − x) = k(1 + x + x2 + x3 + · · · + xn )(1 − x) = k(1 + x + x2 + x3 + · · · + xn )1 − k(1 + x + x2 + x3 + · · · + xn−1 + xn )x = k(1 + x + x2 + x3 + · · · + xn − x − x2 − x3 − · · · − xn − xn+1 ) = k(1 − xn+1 ) so sn (1 − x) = k(1 − xn+1 ) sn = k If |x| < 1, lim xn = 0 so 1 − xn+1 . 1−x n→∞ lim sn = lim k n→∞ n→∞ 1 − xn+1 1 =k . 1−x 1−x Thus, when |x| < 1 the geometric series converges to k/(1 − x). When, for example, k = 1 and x = 1/2: 2n+1 − 1 1 1 − (1/2)n+1 = =2− n sn = n 1 − 1/2 2 2 and ∞ X 1 1 = = 2. n 2 1 − 1/2 n=0 We began the chapter with the series ∞ X 1 , n 2 n=1 namely, the geometric series without the first term 1. Each partial sum of this series is 1 less than the corresponding partial sum for the geometric series, so of course the limit is also one less than the value of the geometric series, that is, ∞ X 1 = 1. 2n n=1 It is not hard to see that the following theorem follows from theorem 11.1.2. P P THEOREM 11.2.2 Suppose that an and bn are convergent series, and c is a constant. Then X X X 1. can is convergent and can = c an 262 Chapter 11 Sequences and Series 2. X X X X (an + bn ) is convergent and (an + bn ) = an + bn . The converses of the two parts of this theorem are subtly different. Suppose that P P an diverges; does can also diverge if c is non-zero? Yes: suppose instead that can P P converges; then by the theorem, (1/c)can converges, but this is the same as an , which P by assumption diverges. Hence can also diverges. Note that we are applying the theorem with an replaced by can and c replaced by (1/c). P P P Now suppose that an and bn diverge; does (an + bn ) also diverge? Now the P P P answer is no: Let an = 1 and bn = −1, so certainly an and bn diverge. But (an + P P P bn ) = (1 + −1) = 0 = 0. Of course, sometimes (an + bn ) will also diverge, for P P P example, if an = bn = 1, then (an + bn ) = (1 + 1) = 2 diverges. In general, the sequence of partial sums sn is harder to understand and analyze than the sequence of terms an , and it is difficult to determine whether series converge and if so to what. Sometimes things are relatively simple, starting with the following. P THEOREM 11.2.3 Proof. Since P If P an converges then lim an = 0. n→∞ an converges, lim sn = L and lim sn−1 = L, because this really says n→∞ n→∞ the same thing but “renumbers” the terms. By theorem 11.1.2, lim (sn − sn−1 ) = lim sn − lim sn−1 = L − L = 0. n→∞ n→∞ n→∞ But sn − sn−1 = (a0 + a1 + a2 + · · · + an ) − (a0 + a1 + a2 + · · · + an−1 ) = an , so as desired lim an = 0. n→∞ This theorem presents an easy divergence test: if given a series P an the limit lim an n→∞ does not exist or has a value other than zero, the series diverges. Note well that the converse is not true: If lim an = 0 then the series does not necessarily converge. n→∞ EXAMPLE 11.2.4 Show that ∞ X n diverges. n+1 n=1 We compute the limit: n = 1 6= 0. n→∞ n + 1 Looking at the first few terms perhaps makes it clear that the series has no chance of converging: 1 2 3 4 + + + +··· 2 3 4 5 lim 11.2 Series 263 will just get larger and larger; indeed, after a bit longer the series starts to look very much like · · · + 1 + 1 + 1 + 1 + · · ·, and of course if we add up enough 1’s we can make the sum as large as we desire. EXAMPLE 11.2.5 ∞ X 1 diverges. Show that n n=1 Here the theorem does not apply: lim 1/n = 0, so it looks like perhaps the series conn→∞ verges. Indeed, if you have the fortitude (or the software) to add up the first 1000 terms you will find that 1000 X 1 ≈ 7.49, n n=1 so it might be reasonable to speculate that the series converges to something in the neighborhood of 10. But in fact the partial sums do go to infinity; they just get big very, very slowly. Consider the following: 1 1 1 1 1 1 1 1 + + >1+ + + =1+ + 2 3 4 2 4 4 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1+ + + + + + + >1+ + + + + + + =1+ + + 2 3 4 5 6 7 8 2 4 4 8 8 8 8 2 2 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 > 1+ + + + +···+ + +···+ = 1+ + + + 1+ + +···+ 2 3 16 2 4 4 8 8 16 16 2 2 2 2 1+ and so on. By swallowing up more and more terms we can always manage to add at least another 1/2 to the sum, and by adding enough of these we can make the partial sums as big as we like. In fact, it’s not hard to see from this pattern that 1+ 1 1 1 n + +···+ n > 1+ , 2 3 2 2 so to make sure the sum is over 100, for example, we’d add up terms until we get to P around 1/2198 , that is, about 4 · 1059 terms. This series, (1/n), is called the harmonic series. Exercises 11.2. 1. Explain why ∞ X n=1 ∞ X n2 diverges. ⇒ +1 2n2 2. Explain why 5 diverges. ⇒ 1/n + 14 2 n=1 3. Explain why ∞ X 3 diverges. ⇒ n n=1 264 Chapter 11 Sequences and Series 4. Compute 5. Compute 6. Compute ∞ X n=0 ∞ X n=0 ∞ X n=0 ∞ X 3 4 − n. ⇒ n (−3) 3 3 4 + n. ⇒ n 2 5 4n+1 . ⇒ 5n 3n+1 . ⇒ 7n+1 n=0 ∞  n X 3 8. Compute . ⇒ 5 n=1 7. Compute 9. Compute 11.3 ∞ X 3n . ⇒ 5n+1 n=1 The Integral Test It is generally quite difficult, often impossible, to determine the value of a series exactly. In many cases it is possible at least to determine whether or not the series converges, and so we will spend most of our time on this problem. If all of the terms an in a series are non-negative, then clearly the sequence of partial sums sn is non-decreasing. This means that if we can show that the sequence of partial sums is bounded, the series must converge. We know that if the series converges, the terms an approach zero, but this does not mean that an ≥ an+1 for every n. Many useful and interesting series do have this property, however, and they are among the easiest to understand. Let’s look at an example. EXAMPLE 11.3.1 ∞ X 1 Show that converges. 2 n n=1 The terms 1/n2 are positive and decreasing, and since lim 1/x2 = 0, the terms 1/n2 x→∞ approach zero. We seek an upper bound for all the partial sums, that is, we want to find a number N so that sn ≤ N for every n. The upper bound is provided courtesy of integration, and is inherent in figure 11.3.1. The figure shows the graph of y = 1/x2 together with some rectangles that lie completely below the curve and that all have base length one. Because the heights of the rectangles are determined by the height of the curve, the areas of the rectangles are 1/12 , 1/22 , 1/32 , and so on—in other words, exactly the terms of the series. The partial sum sn is simply the sum of the areas of the first n rectangles. Because the rectangles all lie between the curve and the x-axis, any sum of rectangle areas is less than the corresponding 11.3 The Integral Test 265 2 1 A= 0 0 ... ... ... ... .. .. .. ... ... ... ... ... ... ... ... ... .... ..... ..... ..... ...... 1 ....... ........ .......... ............. ................. ........................... ................................................... A = 1/4 ................................................................................ 1 Figure 11.3.1 2 3 4 5 Graph of y = 1/x2 with rectangles. area under the curve, and so of course any sum of rectangle areas is less than the area under the entire curve, that is, all the way to infinity. There is a bit of trouble at the left end, where there is an asymptote, but we can work around that easily. Here it is: Z n Z ∞ 1 1 1 1 1 1 dx < 1 + dx = 1 + 1 = 2, sn = 2 + 2 + 2 + · · · + 2 < 1 + 2 1 2 3 n x2 1 x 1 recalling that we computed this improper integral in section 9.7. Since the sequence of partial sums sn is increasing and bounded above by 2, we know that lim sn = L ≤ 2, and n→∞ so the series converges to some number at most 2. In fact, it is possible, though difficult, to show that L = π 2 /6 ≈ 1.6. P We already know that 1/n diverges. What goes wrong if we try to apply this technique to it? Here’s the calculation: Z n Z ∞ 1 1 1 1 1 1 sn = + + + · · · + < 1 + dx < 1 + dx = 1 + ∞. 1 2 3 n x 1 x 1 The problem is that the improper integral doesn’t converge. Note well that this does P not prove that 1/n diverges, just that this particular calculation fails to prove that it P converges. A slight modification, however, allows us to prove in a second way that 1/n diverges. EXAMPLE 11.3.2 Consider a slightly altered version of figure 11.3.1, shown in figure 11.3.2. The rectangles this time are above the curve, that is, each rectangle completely contains the corresponding area under the curve. This means that Z n+1 n+1 1 1 1 1 1 sn = + + + · · · + > dx = ln x = ln(n + 1). 1 2 3 n x 1 1 As n gets bigger, ln(n + 1) goes to infinity, so the sequence of partial sums sn must also go to infinity, so the harmonic series diverges. 266 Chapter 11 Sequences and Series 2 ... ... ... ... ... .. ... ... ... ... ... ... ... .... .... ..... ..... ...... ...... ...... ....... ........ ......... ........... . A = 1 ........................................... ......................... ................................... .................................................. ................................. A = 1/2 1 A = 1/3 0 0 1 Figure 11.3.2 2 3 4 5 Graph of y = 1/x with rectangles. The important fact that clinches this example is that Z n+1 1 dx = ∞, lim n→∞ 1 x which we can rewrite as Z ∞ 1 1 dx = ∞. x So these two examples taken together indicate that we can prove that a series converges or prove that it diverges with a single calculation of an improper integral. This is known as the integral test, which we state as a theorem. Suppose that f (x) > 0 and is decreasing on the infinite interval ∞ X [k, ∞) (for some k ≥ 1) and that an = f (n). Then the series an converges if and only THEOREM 11.3.3 n=1 if the improper integral Z ∞ f (x) dx converges. 1 The two examples we have seen are called p-series; a p-series is any series of the form X 1/np . If p ≤ 0, lim 1/np 6= 0, so the series diverges. For positive values of p we can n→∞ determine precisely which series converge. THEOREM 11.3.4 Proof. A p-series with p > 0 converges if and only if p > 1. We use the integral test; we have already done p = 1, so assume that p 6= 1. Z ∞ D D1−p 1 x1−p 1 = lim dx = lim − . p D→∞ D→∞ x 1−p 1 1−p 1−p 1 If p > 1 then 1 − p < 0 and lim D1−p = 0, so the integral converges. If 0 < p < 1 then D→∞ 1 − p > 0 and lim D D→∞ 1−p = ∞, so the integral diverges. 11.3 EXAMPLE 11.3.5 Show that The Integral Test 267 ∞ X 1 converges. 3 n n=1 We could of course use the integral test, but now that we have the theorem we may simply note that this is a p-series with p > 1. EXAMPLE 11.3.6 We know that if ∞ X Show that ∞ X 5 converges. 4 n n=1 4 1/n converges then n=1 ∞ X ∞ X 5/n4 also converges, by theorem 11.2.2. Since n=1 1/n4 is a convergent p-series, n=1 ∞ X 5/n4 converges also. n=1 EXAMPLE 11.3.7 ∞ X 5 √ diverges. Show that n n=1 ∞ X 1 √ is a p-series with p = 1/2 < 1, it This also follows from theorem 11.2.2: Since n n=1 ∞ X 5 √ . diverges, and so does n n=1 Since it is typically difficult to compute the value of a series exactly, a good approximation is frequently required. In a real sense, a good approximation is only as good as we know it is, that is, while an approximation may in fact be good, it is only valuable in practice if we can guarantee its accuracy to some degree. This guarantee is usually easy to come by for series with decreasing positive terms. EXAMPLE 11.3.8 Approximate X 1/n2 to two decimal places. Referring to figure 11.3.1, if we approximate the sum by N X 1/n2 , the error we make is n=1 the total area of the remaining rectangles, all of which lie under the curve 1/x2 from x = N out to infinity. So we know the true value of the series is larger than the approximation, and no bigger than the approximation plus the area under the curve from N to infinity. Roughly, then, we need to find N so that Z ∞ N 1 dx < 1/100. x2 268 Chapter 11 Sequences and Series We can compute the integral: Z ∞ N 1 1 dx = , 2 x N so N = 100 is a good starting point. Adding up the first 100 terms gives approximately 1.634983900, and that plus 1/100 is 1.644983900, so approximating the series by the value halfway between these will be at most 1/200 = 0.005 in error. The midpoint is 1.639983900, but while this is correct to ±0.005, we can’t tell if the correct two-decimal approximation is 1.63 or 1.64. We need to make N big enough to reduce the guaranteed error, perhaps to around 0.004 to be safe, so we would need 1/N ≈ 0.008, or N = 125. Now the sum of the first 125 terms is approximately 1.636965982, and that plus 0.008 is 1.644965982 and the point halfway between them is 1.640965982. The true value is then 1.640965982±0.004, and all numbers in this range round to 1.64, so 1.64 is correct to two decimal places. We have mentioned that the true value of this series can be shown to be π 2 /6 ≈ 1.644934068 which rounds down to 1.64 (just barely) and is indeed below the upper bound of 1.644965982, again just barely. Frequently approximations will be even better than the “guaranteed” accuracy, but not always, as this example demonstrates. Exercises 11.3. Determine whether each series converges or diverges. ∞ ∞ X X n 1 ⇒ ⇒ 2. 1. 2 +1 π/4 n n n=1 n=1 ∞ X ln n 3. ⇒ n2 n=1 4. ∞ X 1 5. ⇒ n e n=1 6. 7. ∞ X n=2 1 ⇒ n ln n 9. Find an N so that 8. ∞ X n=1 ∞ X n=1 ∞ X n=2 1 ⇒ n2 + 1 n ⇒ en 1 ⇒ n(ln n)2 N N ∞ X X X 1 1 1 is between and + 0.005. ⇒ 4 4 4 n n n n=1 n=1 n=1 10. Find an N so that ∞ N N X X X 1 1 1 is between and + 10−4 . ⇒ n n n e e e n=0 n=0 n=0 11. Find an N so that ∞ N N X X X ln n ln n ln n is between and + 0.005. ⇒ 2 2 2 n n n n=1 n=1 n=1 12. Find an N so that ∞ X n=2 N N X X 1 1 1 is between and + 0.005. ⇒ 2 2 n(ln n) n(ln n) n(ln n)2 n=2 n=2 11.4 11.4 Alternating Series 269 Alternating Series Next we consider series with both positive and negative terms, but in a regular pattern: they alternate, as in the alternating harmonic series for example: ∞ X (−1)n−1 1 −1 1 −1 1 1 1 1 = + + + + ··· = − + − + ···. n 1 2 3 4 1 2 3 4 n=1 In this series the sizes of the terms decrease, that is, |an | forms a decreasing sequence, but this is not required in an alternating series. As with positive term series, however, when the terms do have decreasing sizes it is easier to analyze the series, much easier, in fact, than positive term series. Consider pictorially what is going on in the alternating harmonic series, shown in figure 11.4.1. Because the sizes of the terms an are decreasing, the partial sums s1 , s3 , s5 , and so on, form a decreasing sequence that is bounded below by s2 , so this sequence must converge. Likewise, the partial sums s2 , s4 , s6 , and so on, form an increasing sequence that is bounded above by s1 , so this sequence also converges. Since all the even numbered partial sums are less than all the odd numbered ones, and since the “jumps” (that is, the ai terms) are getting smaller and smaller, the two sequences must converge to the same value, meaning the entire sequence of partial sums s1 , s2 , s3 , . . . converges as well. a2 = − 1 2 ..................................................................................................................................................................................................................................... a a 6 ............................................................................. 4 ...................................................................................................................... ...... s2 = 1 4 1 2 s4 s6 s3 s5 a 1 = s1 = a 1 ..............................................5 .................................................. ........................................................................................................................................................... a3 Figure 11.4.1 The alternating harmonic series. There’s nothing special about the alternating harmonic series—the same argument works for any alternating sequence with decreasing size terms. The alternating series test is worth calling a theorem. Suppose that {an }∞ n=1 is a non-increasing sequence of positive ∞ X numbers and lim an = 0. Then the alternating series (−1)n−1 an converges. THEOREM 11.4.1 n→∞ n=1 Proof. The odd numbered partial sums, s1 , s3 , s5 , and so on, form a non-increasing sequence, because s2k+3 = s2k+1 − a2k+2 + a2k+3 ≤ s2k+1 , since a2k+2 ≥ a2k+3 . This 270 Chapter 11 Sequences and Series sequence is bounded below by s2 , so it must converge, say lim s2k+1 = L. Likewise, k→∞ the partial sums s2 , s4 , s6 , and so on, form a non-decreasing sequence that is bounded above by s1 , so this sequence also converges, say lim s2k = M . Since lim an = 0 and n→∞ k→∞ s2k+1 = s2k + a2k+1 , L = lim s2k+1 = lim (s2k + a2k+1 ) = lim s2k + lim a2k+1 = M + 0 = M, k→∞ k→∞ k→∞ k→∞ so L = M , the two sequences of partial sums converge to the same limit, and this means the entire sequence of partial sums also converges to L. Another useful fact is implicit in this discussion. Suppose that ∞ X L= (−1)n−1 an n=1 and that we approximate L by a finite part of this sum, say L≈ N X (−1)n−1 an . n=1 Because the terms are decreasing in size, we know that the true value of L must be between this approximation and the next one, that is, between N X (−1) n−1 an and n=1 N+1 X (−1)n−1 an . n=1 Depending on whether N is odd or even, the second will be larger or smaller than the first. EXAMPLE 11.4.2 Approximate the alternating harmonic series to one decimal place. We need to go roughly to the point at which the next term to be added or subtracted is 1/10. Adding up the first nine and the first ten terms we get approximately 0.746 and 0.646. These are 1/10 apart, but it is not clear how the correct value would be rounded. It turns out that we are able to settle the question by computing the sums of the first eleven and twelve terms, which give 0.737 and 0.653, so correct to one place the value is 0.7. We have considered alternating series with first index 1, and in which the first term is positive, but a little thought shows this is not crucial. The same test applies to any similar ∞ ∞ ∞ X X X (−1)n an , etc. (−1)n an , (−1)n an , series, such as n=0 n=1 n=17 11.5 Comparison Tests 271 Exercises 11.4. Determine whether the following series converge or diverge. ∞ ∞ X X (−1)n−1 (−1)n−1 √ ⇒ 2. 1. ⇒ 2n + 5 n−3 n=4 n=1 3. ∞ X (−1) n−1 n=1 n ⇒ 3n − 2 5. Approximate 6. Approximate ∞ X (−1)n−1 n=1 ∞ X (−1)n−1 n=1 11.5 4. ∞ X (−1)n−1 n=1 ln n ⇒ n 1 to two decimal places. ⇒ n3 1 to two decimal places. ⇒ n4 Comparison Tests As we begin to compile a list of convergent and divergent series, new ones can sometimes be analyzed by comparing them to ones that we already understand. EXAMPLE 11.5.1 Does ∞ X 1 converge? 2 ln n n n=2 The obvious first approach, based on what we know, is the integral test. Unfortunately, we can’t compute the required antiderivative. But looking at the series, it would appear that it must converge, because the terms we are adding are smaller than the terms of a p-series, that is, 1 1 < 2, 2 n ln n n when n ≥ 3. Since adding up the terms 1/n2 doesn’t get “too big”, the new series “should” also converge. Let’s make this more precise. ∞ ∞ X X 1 1 The series converges if and only if converges—all we’ve done is 2 2 n ln n n ln n n=2 n=3 ∞ X 1 dropped the initial term. We know that converges. Looking at two typical partial n2 n=3 sums: sn = 32 1 1 1 1 1 1 1 1 + 2 + 2 +···+ 2 < 2 + 2 + 2 + · · · + 2 = tn . ln 3 4 ln 4 5 ln 5 n ln n 3 4 5 n Since the p-series converges, say to L, and since the terms are positive, tn < L. Since the terms of the new series are positive, the sn form an increasing sequence and sn < tn < L for all n. Hence the sequence {sn } is bounded and so converges. 272 Chapter 11 Sequences and Series Sometimes, even when the integral test applies, comparison to a known series is easier, so it’s generally a good idea to think about doing a comparison before doing the integral test. EXAMPLE 11.5.2 Does ∞ X | sin n| converge? 2 n n=1 We can’t apply the integral test here, because the terms of this series are not decreasing. Just as in the previous example, however, | sin n| 1 ≤ , n2 n2 because | sin n| ≤ 1. Once again the partial sums are non-decreasing and bounded above X by 1/n2 = L, so the new series converges. Like the integral test, the comparison test can be used to show both convergence and divergence. In the case of the integral test, a single calculation will confirm whichever is the case. To use the comparison test we must first have a good idea as to convergence or divergence and pick the sequence for comparison accordingly. EXAMPLE 11.5.3 Does ∞ X n=2 √ 1 n2 − 3 converge? We observe that the −3 should have little effect compared to the n2 inside the square √ root, and therefore guess that the terms are enough like 1/ n2 = 1/n that the series should diverge. We attempt to show this by comparison to the harmonic series. We note that 1 1 1 √ >√ = , 2 2 n n −3 n so that sn = √ 1 22 − 3 +√ 1 32 − 3 +···+ √ 1 n2 − 3 > 1 1 1 + + · · · + = tn , 2 3 n where tn is 1 less than the corresponding partial sum of the harmonic series (because we start at n = 2 instead of n = 1). Since lim tn = ∞, lim sn = ∞ as well. n→∞ n→∞ So the general approach is this: If you believe that a new series is convergent, attempt to find a convergent series whose terms are larger than the terms of the new series; if you believe that a new series is divergent, attempt to find a divergent series whose terms are smaller than the terms of the new series. 11.5 EXAMPLE 11.5.4 Does ∞ X n=1 √ 1 n2 + 3 Comparison Tests 273 converge? Just as in the last example, we guess that this is very much like the harmonic series and so diverges. Unfortunately, 1 1 √ < , 2 n n +3 so we can’t compare the series directly to the harmonic series. A little thought leads us to √ 1 n2 +3 >√ n2 1 1 = , 2 2n + 3n P P P so if 1/(2n) diverges then the given series diverges. But since 1/(2n) = (1/2) 1/n, theorem 11.2.2 implies that it does indeed diverge. For reference we summarize the comparison test in a theorem. THEOREM 11.5.5 Suppose that an and bn are non-negative for all n and that an ≤ bn when n ≥ N , for some N . ∞ ∞ X X If bn converges, so does an . n=0 If ∞ X n=0 an diverges, so does n=0 ∞ X bn . n=0 Exercises 11.5. Determine whether the series converge or diverge. ∞ ∞ X X 1 1 ⇒ 2. ⇒ 1. 2 2 2n + 3n + 5 2n + 3n − 5 n=2 n=1 3. 5. 7. 9. ∞ X n=1 ∞ X n=1 ∞ X n=1 ∞ X n=1 1 ⇒ 2 2n − 3n − 5 4. 2 3n + 4 ⇒ 2n2 + 3n + 5 6. ln n ⇒ n3 8. n 3 ⇒ 2n + 5n 10. ∞ X n=1 ∞ X n=1 ∞ X n=2 ∞ X n=1 3n + 4 ⇒ + 3n + 5 2n2 ln n ⇒ n 1 ⇒ ln n 3n ⇒ 2n + 3n 274 Chapter 11 Sequences and Series 11.6 Absolute Convergen e P Roughly speaking there are two ways for a series to converge: As in the case of 1/n2 , the individual terms get small very quickly, so that the sum of all of them stays finite, or, X P as in the case of (−1)n−1 /n, the terms don’t get small fast enough ( 1/n diverges), but a mixture of positive and negative terms provides enough cancellation to keep the sum finite. You might guess from what we’ve seen that if the terms get small fast enough to do the job, then whether or not some terms are negative and some positive the series converges. THEOREM 11.6.1 If ∞ X n=0 Proof. Now |an | converges, then ∞ X an converges. n=0 Note that 0 ≤ an +|an | ≤ 2|an | so by the comparison test ∞ X n=0 (an + |an |) − ∞ X n=0 |an | = ∞ X n=0 ∞ X n=0 an + |an | − |an | = (an +|an |) converges. ∞ X an n=0 converges by theorem 11.2.2. P So given a series an with both positive and negative terms, you should first ask P whether |an | converges. This may be an easier question to answer, because we have P tests that apply specifically to series with non-negative terms. If |an | converges then P P you know that an converges as well. If |an | diverges then it still may be true that P an converges—you will have to do more work to decide the question. Another way to P P think of this result is: it is (potentially) easier for an to converge than for |an | to converge, because the latter series cannot take advantage of cancellation. P P P If |an | converges we say that an converges absolutely; to say that an converges absolutely is to say that any cancellation that happens to come along is not really needed, as the terms already get small so fast that convergence is guaranteed by that alone. If P P P an converges but |an | does not, we say that an converges conditionally. For ∞ ∞ X X 1 1 (−1)n−1 converges conditionally. (−1)n−1 2 converges absolutely, while example n n n=1 n=1 EXAMPLE 11.6.2 ∞ X sin n Does converge? n2 n=2 11.7 In example 11.5.2 we saw that The Ratio and Root Tests 275 ∞ X | sin n| converges, so the given series converges abso2 n n=2 lutely. EXAMPLE 11.6.3 Does ∞ X (−1)n n=0 3n + 4 converge? + 3n + 5 2n2 ∞ X 3 3n + 4 diverges by comparison to , so if the Taking the absolute value, 2 2n + 3n + 5 10n n=1 n=0 ∞ X series converges it does so conditionally. It is true that lim (3n + 4)/(2n2 + 3n + 5) = 0, n→∞ so to apply the alternating series test we need to know whether the terms are decreasing. If we let f (x) = (3x + 4)/(2x2 + 3x + 5) then f ′ (x) = −(6x2 + 16x − 3)/(2x2 + 3x + 5)2 , and it is not hard to see that this is negative for x ≥ 1, so the series is decreasing and by the alternating series test it converges. Exercises 11.6. Determine whether each series converges absolutely, converges conditionally, or diverges. ∞ ∞ X X 3n2 + 4 1 n−1 n−1 ⇒ 2. ⇒ (−1) 1. (−1) 2n2 + 3n + 5 2n2 + 3n + 5 n=1 n=1 3. 5. 7. ∞ X (−1) n−1 ln n n=1 ∞ X (−1)n n=2 ∞ X (−1)n n=0 11.7 n ⇒ 4. 1 ⇒ ln n 6. n 2n 3 ⇒ + 3n 8. ∞ X (−1)n−1 n=1 ∞ X (−1)n n=0 ∞ X 3n ⇒ 2n + 5n (−1)n−1 n=1 ln n ⇒ n3 arctan n ⇒ n The Ratio and Root Tests ∞ X n5 converge? It is possible, but a bit unpleasant, to approach this Does the series 5n n=0 with the integral test or the comparison test, but there is an easier way. Consider what happens as we move from one term to the next in this series: ···+ n5 (n + 1)5 + +··· 5n 5n+1 The denominator goes up by a factor of 5, 5n+1 = 5 · 5n , but the numerator goes up by much less: (n + 1)5 = n5 + 5n4 + 10n3 + 10n2 + 5n + 1, which is much less than 5n5 when n is large, because 5n4 is much less than n5 . So we might guess that in the long run it 276 Chapter 11 Sequences and Series begins to look as if each term is 1/5 of the previous term. We have seen series that behave like this: ∞ X 1 5 = , 5n 4 n=0 a geometric series. So we might try comparing the given series to some variation of this geometric series. This is possible, but a bit messy. We can in effect do the same thing, but bypass most of the unpleasant work. The key is to notice that 1 1 an+1 (n + 1)5 5n (n + 1)5 1 = lim = lim =1· = . n+1 5 5 n→∞ an n→∞ 5 n→∞ n n 5 5 5 lim This is really just what we noticed above, done a bit more officially: in the long run, each term is one fifth of the previous term. Now pick some number between 1/5 and 1, say 1/2. Because 1 an+1 = , lim n→∞ an 5 then when n is big enough, say n ≥ N for some N , 1 an+1 < an 2 and an+1 < an . 2 So aN+1 < aN /2, aN+2 < aN+1 /2 < aN /4, aN+3 < aN+2 /2 < aN+1 /4 < aN /8, and so on. The general form is aN+k < aN /2k . So if we look at the series ∞ X k=0 aN+k = aN + aN+1 + aN+2 + aN+3 + · · · + aN+k + · · · , its terms are less than or equal to the terms of the sequence ∞ X aN aN aN aN aN aN + + + +···+ k +··· = = 2aN . 2 4 8 2 2k k=0 So by the comparison test, ∞ X k=0 aN+k converges, and this means that ∞ X an converges, n=0 since we’ve just added the fixed number a0 + a1 + · · · + aN−1 . Under what circumstances could we do this? What was crucial was that the limit of an+1 /an , say L, was less than 1 so that we could pick a value r so that L < r < 1. The P fact that L < r (1/5 < 1/2 in our example) means that we can compare the series an P n P n to r , and the fact that r < 1 guarantees that r converges. That’s really all that is 11.7 The Ratio and Root Tests 277 required to make the argument work. We also made use of the fact that the terms of the series were positive; in general we simply consider the absolute values of the terms and we end up testing for absolute convergence. THEOREM 11.7.1 The Ratio Test Suppose that lim |an+1 /an | = L. If L < 1 n→∞ P the series an converges absolutely, if L > 1 the series diverges, and if L = 1 this test gives no information. Proof. The example above essentially proves the first part of this, if we simply replace 1/5 by L and 1/2 by r. Suppose that L > 1, and pick r so that 1 < r < L. Then for n ≥ N , for some N , |an+1 | > r and |an+1 | > r|an |. |an | This implies that |aN+k | > r k |aN |, but since r > 1 this means that lim |aN+k | = 6 0, which k→∞ means also that lim an 6= 0. By the divergence test, the series diverges. n→∞ To see that we get no information when L = 1, we need to exhibit two series with P P L = 1, one that converges and one that diverges. It is easy to see that 1/n2 and 1/n do the job. EXAMPLE 11.7.2 The ratio test is particularly useful for series involving the factorial ∞ X function. Consider 5n /n!. n=0 5n+1 n! 5n+1 1 n! = lim = lim 5 = 0. n n n→∞ (n + 1)! 5 n→∞ 5 (n + 1)! n→∞ (n + 1) lim Since 0 < 1, the series converges. A similar argument, which we will not do, justifies a similar test that is occasionally easier to apply. THEOREM 11.7.3 The Root Test Suppose that lim |an |1/n = L. If L < 1 the n→∞ P series an converges absolutely, if L > 1 the series diverges, and if L = 1 this test gives no information. The proof of the root test is actually easier than that of the ratio test, and is a good exercise. EXAMPLE 11.7.4 ∞ X 5n . Analyze n n n=0 278 Chapter 11 Sequences and Series The ratio test turns out to be a bit difficult on this series (try it). Using the root test: lim n→∞  5n nn 1/n (5n )1/n 5 = 0. = lim n 1/n n→∞ (n ) n→∞ n = lim Since 0 < 1, the series converges. The root test is frequently useful when n appears as an exponent in the general term of the series. Exercises 11.7. 1. Compute lim |an+1 /an | for the series n→∞ P 1/n2 . P 2. Compute lim |an+1 /an | for the series 1/n. n→∞ P 3. Compute lim |an |1/n for the series 1/n2 . n→∞ P 4. Compute lim |an |1/n for the series 1/n. n→∞ Determine whether the series converge. ∞ X 3n 5. (−1)n n ⇒ 5 n=0 6. ∞ X n! ⇒ nn n=1 7. ∞ X n5 ⇒ nn n=1 8. ∞ X (n!)2 ⇒ nn n=1 9. Prove theorem 11.7.3, the root test. 11.8 Power Series Recall that we were able to analyze all geometric series “simultaneously” to discover that ∞ X n=0 kxn = k , 1−x if |x| < 1, and that the series diverges when |x| ≥ 1. At the time, we thought of x as an unspecified constant, but we could just as well think of it as a variable, in which case the 11.8 series ∞ X Power Series 279 kxn n=0 is a function, namely, the function k/(1 − x), as long as |x| < 1. While k/(1 − x) is a reaP n sonably easy function to deal with, the more complicated kx does have its attractions: it appears to be an infinite version of one of the simplest function types—a polynomial. This leads naturally to the questions: Do other functions have representations as series? Is there an advantage to viewing them in this way? The geometric series has a special feature that makes it unlike a typical polynomial— the coefficients of the powers of x are the same, namely k. We will need to allow more general coefficients if we are to get anything other than the geometric series. DEFINITION 11.8.1 A power series has the form ∞ X a n xn , n=0 with the understanding that an may depend on n but not on x. EXAMPLE 11.8.2 ratio test: ∞ X xn is a power series. We can investigate convergence using the n n=1 n |x|n+1 n = lim |x| = |x|. n→∞ n→∞ n + 1 |x|n n+1 lim Thus when |x| < 1 the series converges and when |x| > 1 it diverges, leaving only two values in doubt. When x = 1 the series is the harmonic series and diverges; when x = −1 it is the alternating harmonic series (actually the negative of the usual alternating harmonic series) ∞ X xn and converges. Thus, we may think of as a function from the interval [−1, 1) to n n=1 the real numbers. A bit of thought reveals that the ratio test applied to a power series will always have the same nice form. In general, we will compute |an+1 ||x|n+1 |an+1 | |an+1 | = lim |x| = |x| lim = L|x|, n n→∞ n→∞ n→∞ |an | |an ||x| |an | lim assuming that lim |an+1 |/|an | exists. Then the series converges if L|x| < 1, that is, if |x| < 1/L, and diverges if |x| > 1/L. Only the two values x = ±1/L require further 280 Chapter 11 Sequences and Series investigation. Thus the series will definitely define a function on the interval (−1/L, 1/L), and perhaps will extend to one or both endpoints as well. Two special cases deserve mention: if L = 0 the limit is 0 no matter what value x takes, so the series converges for all x and the function is defined for all real numbers. If L = ∞, then for any non-zero value of x the limit is infinite, so the series converges only when x = 0. The value 1/L is called the radius of convergence of the series, and the interval on which the series converges is the interval of convergence. Consider again the geometric series, ∞ X n=0 xn = 1 . 1−x Whatever benefits there might be in using the series form of this function are only available to us when x is between −1 and 1. Frequently we can address this shortcoming by modifying the power series slightly. Consider this series: n ∞  ∞ X X 3 1 x+2 (x + 2)n = = = , 3n 3 1−x 1 − x+2 3 n=0 n=0 because this is just a geometric series with x replaced by (x + 2)/3. Multiplying both sides by 1/3 gives ∞ X 1 (x + 2)n = , n+1 3 1−x n=0 the same function as before. For what values of x does this series converge? Since it is a geometric series, we know that it converges when |x + 2|/3 < 1 |x + 2| < 3 −3 < x + 2 < 3 −5 < x < 1. So we have a series representation for 1/(1 − x) that works on a larger interval than before, at the expense of a somewhat more complicated series. The endpoints of the interval of convergence now are −5 and 1, but note that they can be more compactly described as −2 ± 3. We say that 3 is the radius of convergence, and we now say that the series is centered at −2. 11.9 DEFINITION 11.8.3 Calculus with Power Series 281 A power series centered at a has the form ∞ X n=0 an (x − a)n , with the understanding that an may depend on n but not on x. Exercises 11.8. Find the radius and interval of convergence for each series. In exercises 3 and 4, do not attempt to determine whether the endpoints are in the interval of convergence. ∞ ∞ X X xn ⇒ 1. nxn ⇒ 2. n! n=0 n=0 3. ∞ X n! n x ⇒ n n n=1 4. ∞ X n! (x − 2)n ⇒ n n n=1 5. ∞ X (n!)2 (x − 2)n ⇒ n n n=1 6. ∞ X (x + 5)n ⇒ n(n + 1) n=1 11.9 Cal ulus with Power Series Now we know that some functions can be expressed as power series, which look like infinite polynomials. Since calculus, that is, computation of derivatives and antiderivatives, is easy for polynomials, the obvious question is whether the same is true for infinite series. The answer is yes: THEOREM 11.9.1 Suppose the power series f (x) = ∞ X n=0 an (x − a)n has radius of convergence R. Then ′ f (x) = ∞ X n=0 Z nan (x − a)n−1 , ∞ X an (x − a)n+1 , f (x) dx = C + n + 1 n=0 and these two series have radius of convergence R as well. 282 Chapter 11 Sequences and Series Starting with the geometric series: EXAMPLE 11.9.2 ∞ X 1 = xn 1 − x n=0 Z ∞ X 1 1 dx = − ln |1 − x| + C = xn+1 1−x n+1 n=0 ln |1 − x| = C + ∞ X n=0 − 1 xn+1 n+1 We can figure out what C is: When x = 0, this becomes ln |1| = C + 0, so C = 0, and P∞ 1 xn+1 , when |x| < 1. The series does not converge when x = 1 ln |1 − x| = n=0 − n+1 but does converge when x = −1 or 1 − x = 2. The interval of convergence is [−1, 1), or 0 < 1 − x ≤ 2, so we can use the series to represent ln(x) when 0 < x ≤ 2. For example ln(3/2) = ln(1 − −1/2) = ∞ X (−1)n n=0 1 1 n+1 n+12 and so 1 1 1 1 1 909 1 1 − + − + − + = ≈ 0.406. 2 8 24 64 160 384 896 2240 Because this is an alternating series with decreasing terms, we know that the true value is between 909/2240 and 909/2240 − 1/2048 = 29053/71680 ≈ .4053, so correct to two decimal places the value is 0.41. What about ln(9/4)? Since 9/4 is larger than 2 we cannot use the series directly, but ln(3/2) ≈ ln(9/4) = ln((3/2)2 ) = 2 ln(3/2) ≈ 0.82, so in fact we get a lot more from this one calculation than first meets the eye. To estimate the true value accurately we actually need to be a bit more careful. When we multiply by two we know that the true value is between 0.8106 and 0.812, so rounded to two decimal places the true value is 0.81. Exercises 11.9. 1. Find a series representation for ln 2. ⇒ 2. Find a power series representation for 1/(1 − x)2 . ⇒ 3. Find a power series representation for 2/(1 − x)3 . ⇒ 4. Find a power series representation for 1/(1 − x)3 . What is the radius of convergence? ⇒ Z 5. Find a power series representation for ln(1 − x) dx. ⇒ 11.10 11.10 Taylor Series 283 Taylor Series We have seen that some functions can be represented as series, which may give valuable information about the function. So far, we have seen only those examples that result from manipulation of our one fundamental example, the geometric series. We would like to start with a given function and produce a series to represent it, if possible. ∞ X an xn on some interval of convergence. Then we know that Suppose that f (x) = n=0 we can compute derivatives of f by taking derivatives of the terms of the series. Let’s look at the first few in general: ′ f (x) = ∞ X nan xn−1 = a1 + 2a2 x + 3a3 x2 + 4a4 x3 + · · · ∞ X n(n − 1)an xn−2 = 2a2 + 3 · 2a3 x + 4 · 3a4 x2 + · · · ∞ X n(n − 1)(n − 2)an xn−3 = 3 · 2a3 + 4 · 3 · 2a4 x + · · · n=1 f ′′ (x) = n=2 ′′′ f (x) = n=3 By examining these it’s not hard to discern the general pattern. The kth derivative must be ∞ X (k) f (x) = n(n − 1)(n − 2) · · · (n − k + 1)an xn−k n=k = k(k − 1)(k − 2) · · · (2)(1)ak + (k + 1)(k) · · · (2)ak+1 x + + (k + 2)(k + 1) · · · (3)ak+2 x2 + · · · We can shrink this quite a bit by using factorial notation: f (k) (x) = ∞ X n=k n! (k + 2)! an xn−k = k!ak + (k + 1)!ak+1 x + ak+2 x2 + · · · (n − k)! 2! Now substitute x = 0: f (k) (0) = k!ak + ∞ X n=k+1 n! an 0n−k = k!ak , (n − k)! and solve for ak : ak = f (k) (0) . k! Note the special case, obtained from the series for f itself, that gives f (0) = a0 . 284 Chapter 11 Sequences and Series So if a function f can be represented by a series, we know just what series it is. Given a function f , the series ∞ X f (n) (0) n x n! n=0 is called the Maclaurin series for f . EXAMPLE 11.10.1 Find the Maclaurin series for f (x) = 1/(1 − x). We need to compute the derivatives of f (and hope to spot a pattern). f (x) = (1 − x)−1 f ′ (x) = (1 − x)−2 f ′′ (x) = 2(1 − x)−3 f ′′′ (x) = 6(1 − x)−4 f (4) (x) = 4!(1 − x)−5 .. . f (n) (x) = n!(1 − x)−n−1 So n!(1 − 0)−n−1 f (n) (0) = =1 n! n! and the Maclaurin series is ∞ X n=0 n 1·x = ∞ X xn , n=0 the geometric series. A warning is in order here. Given a function f we may be able to compute the Maclaurin series, but that does not mean we have found a series representation for f . We still need to know where the series converges, and if, where it converges, it converges to f (x). While for most commonly encountered functions the Maclaurin series does indeed converge to f on some interval, this is not true of all functions, so care is required. As a practical matter, if we are interested in using a series to approximate a function, we will need some finite number of terms of the series. Even for functions with messy derivatives we can compute these using computer software like Sage. If we want to know the whole series, that is, a typical term in the series, we need a function whose derivatives fall into a pattern that we can discern. A few of the most important functions are fortunately very easy. 11.10 Taylor Series 285 EXAMPLE 11.10.2 Find the Maclaurin series for sin x. The derivatives are quite easy: f ′ (x) = cos x, f ′′ (x) = − sin x, f ′′′ (x) = − cos x, f (4) (x) = sin x, and then the pattern repeats. We want to know the derivatives at zero: 1, 0, −1, 0, 1, 0, −1, 0,. . . , and so the Maclaurin series is x− ∞ X x3 x5 x2n+1 + −··· = (−1)n . 3! 5! (2n + 1)! n=0 We should always determine the radius of convergence: |x|2 |x|2n+3 (2n + 1)! = lim = 0, lim n→∞ (2n + 3)(2n + 2) n→∞ (2n + 3)! |x|2n+1 so the series converges for every x. Since it turns out that this series does indeed converge to sin x everywhere, we have a series representation for sin x for every x. Sometimes the formula for the nth derivative of a function f is difficult to discover, but a combination of a known Maclaurin series and some algebraic manipulation leads easily to the Maclaurin series for f . EXAMPLE 11.10.3 Find the Maclaurin series for x sin(−x). To get from sin x to x sin(−x) we substitute −x for x and then multiply by x. We can do the same thing to the series for sin x: x ∞ X n=0 (−1) n (−x) 2n+1 (2n + 1)! =x ∞ X n (−1) (−1) n=0 2n+1 ∞ X x2n+1 x2n+2 = (−1)n+1 . (2n + 1)! n=0 (2n + 1)! As we have seen, a general power series can be centered at a point other than zero, and the method that produces the Maclaurin series can also produce such series. EXAMPLE 11.10.4 Find a series centered at −2 for 1/(1 − x). ∞ X If the series is an (x + 2)n then looking at the kth derivative: n=0 k!(1 − x) −k−1 = ∞ X n=k n! an (x + 2)n−k (n − k)! and substituting x = −2 we get k!3−k−1 = k!ak and ak = 3−k−1 = 1/3k+1 , so the series is ∞ X (x + 2)n . n+1 3 n=0 We’ve already seen this, on page 280. 286 Chapter 11 Sequences and Series Such a series is called the Taylor series for the function, and the general term has the form f (n) (a) (x − a)n . n! A Maclaurin series is simply a Taylor series with a = 0. Exercises 11.10. For each function, find the Maclaurin series or Taylor series centered at a, and the radius of convergence. 1. cos x ⇒ 2. ex ⇒ 3. 1/x, a = 5 ⇒ 4. ln x, a = 1 ⇒ 5. ln x, a = 2 ⇒ 6. 1/x2 , a = 1 ⇒ √ 7. 1/ 1 − x ⇒ 8. Find the first four terms of the Maclaurin series for tan x (up to and including the x3 term). ⇒ 9. Use a combination of Maclaurin series and algebraic manipulation to find a series centered at zero for x cos(x2 ). ⇒ 10. Use a combination of Maclaurin series and algebraic manipulation to find a series centered at zero for xe−x . ⇒ 11.11 Taylor's Theorem One of the most important uses of infinite series is the potential for using an initial portion of the series for f to approximate f . We have seen, for example, that when we add up the first n terms of an alternating series with decreasing terms that the difference between this and the true value is at most the size of the next term. A similar result is true of many Taylor series. THEOREM 11.11.1 Suppose that f is defined on some open interval I around a and suppose f (N+1) (x) exists on this interval. Then for each x 6= a in I there is a value z between x and a so that N X f (N+1) (z) f (n) (a) (x − a)n + (x − a)N+1 . f (x) = n! (N + 1)! n=0 11.11 Taylor’s Theorem 287 Proof. The proof requires some cleverness to set up, but then the details are quite elementary. We want to define a function F (t). Start with the equation F (t) = N X f (n) (t) (x − t)n + B(x − t)N+1 . n! n=0 Here we have replaced a by t in the first N + 1 terms of the Taylor series, and added a carefully chosen term on the end, with B to be determined. Note that we are temporarily keeping x fixed, so the only variable in this equation is t, and we will be interested only in t between a and x. Now substitute t = a: N X f (n) (a) F (a) = (x − a)n + B(x − a)N+1 . n! n=0 Set this equal to f (x): f (x) = N X f (n) (a) (x − a)n + B(x − a)N+1 . n! n=0 Since x 6= a, we can solve this for B, which is a “constant”—it depends on x and a but those are temporarily fixed. Now we have defined a function F (t) with the property that F (a) = f (x). Consider also F (x): all terms with a positive power of (x − t) become zero when we substitute x for t, so we are left with F (x) = f (0) (x)/0! = f (x). So F (t) is a function with the same value on the endpoints of the interval [a, x]. By Rolle’s theorem (6.5.1), we know that there is a value z ∈ (a, x) such that F ′ (z) = 0. Let’s look at F ′ (t). Each term in F (t), except the first term and the extra term involving B, is a product, so to take the derivative we use the product rule on each of these terms. It will help to write out the first few terms of the definition: f (2) (t) f (3) (t) f (1) (t) (x − t)1 + (x − t)2 + (x − t)3 + · · · 1! 2! 3! (N) f (t) (x − t)N + B(x − t)N+1 . + N! F (t) = f (t) + 288 Chapter 11 Sequences and Series Now take the derivative: F (t) = f (t) +  f (2) (t) f (1) (t) 0 (x − t) (−1) + (x − t)1 1! 1!  +  f (2) (t) f (3) (t) (x − t)1 (−1) + (x − t)2 1! 2!  +  f (3) (t) f (4) (t) (x − t)2 (−1) + (x − t)3 2! 3!  +  f (N) (t) f (N+1) (t) (x − t)N−1 (−1) + (x − t)N (N − 1)! N! ′ ′ + ...+  + B(N + 1)(x − t)N (−1). Now most of the terms in this expression cancel out, leaving just F ′ (t) = f (N+1) (t) (x − t)N + B(N + 1)(x − t)N (−1). N! At some z, F ′ (z) = 0 so f (N+1) (z) (x − z)N + B(N + 1)(x − z)N (−1) N! f (N+1) (z) B(N + 1)(x − z)N = (x − z)N N! f (N+1) (z) . B= (N + 1)! 0= Now we can write F (t) = N X f (N+1) (z) f (n) (t) (x − t)n + (x − t)N+1 . n! (N + 1)! n=0 Recalling that F (a) = f (x) we get f (x) = N X f (n) (a) f (N+1) (z) (x − a)n + (x − a)N+1 , n! (N + 1)! n=0 which is what we wanted to show. It may not be immediately obvious that this is particularly useful; let’s look at some examples. 11.11 Taylor’s Theorem 289 EXAMPLE 11.11.2 Find a polynomial approximation for sin x accurate to ±0.005. From Taylor’s theorem: N X f (N+1) (z) f (n) (a) (x − a)n + (x − a)N+1 . sin x = n! (N + 1)! n=0 What can we say about the size of the term f (N+1) (z) (x − a)N+1 ? (N + 1)! Every derivative of sin x is ± sin x or ± cos x, so |f (N+1) (z)| ≤ 1. The factor (x − a)N+1 is a bit more difficult, since x − a could be quite large. Let’s pick a = 0 and |x| ≤ π/2; if we can compute sin x for x ∈ [−π/2, π/2], we can of course compute sin x for all x. We need to pick N so that xN+1 < 0.005. (N + 1)! Since we have limited x to [−π/2, π/2], 2N+1 xN+1 < . (N + 1)! (N + 1)! The quantity on the right decreases with increasing N , so all we need to do is find an N so that 2N+1 < 0.005. (N + 1)! A little trial and error shows that N = 8 works, and in fact 29 /9! < 0.0015, so 8 X f (n) (0) n x ± 0.0015 sin x = n! n=0 =x− x3 x5 x7 + − ± 0.0015. 6 120 5040 Figure 11.11.1 shows the graphs of sin x and and the approximation on [0, 3π/2]. As x gets larger, the approximation heads to negative infinity very quickly, since it is essentially acting like −x7 . 290 Chapter 11 Sequences and Series 1 0 −1 −2 −3 .......................................... .......................... ............... .............. ............. ............ ............. . . . . . . . . . . ............... .... . . . . . . . . ........... . .. ................... .......... ................. .................. ....... ............ ....... ............ ... ....... ....... .......................................... ...... ...... ...... ..... ..... ..... ..... ..... ..... ..... .... ... ... .... ... 1 2 3 4 5 −4 −5 Figure 11.11.1 sin x and a polynomial approximation. (AP) We can extract a bit more information from this example. If we do not limit the value of x, we still have xN+1 f (N+1) (z) N+1 ≤ x (N + 1)! (N + 1)! so that sin x is represented by N X f (n) (0) n xN+1 . x ± n! (N + 1)! n=0 If we can show that lim N→∞ for each x then xN+1 =0 (N + 1)! ∞ ∞ X f (n) (0) n X x2n+1 sin x = x = (−1)n , n! (2n + 1)! n=0 n=0 that is, the sine function is actually equal to its Maclaurin series for all x. How can we prove that the limit is zero? Suppose that N is larger than |x|, and let M be the largest integer less than |x| (if M = 0 the following is even easier). Then |xN+1 | |x| |x| |x| |x| |x| |x| |x| |x| = ··· ··· (N + 1)! N +1 N N −1 M +1M M −1 2 1 ≤ = |x| |x| |x| |x| |x| · 1 · 1···1 · ··· N +1 M M −1 2 1 |x| |x|M . N + 1 M! The quantity |x|M /M ! is a constant, so |x| |x|M =0 lim N→∞ N + 1 M ! 11.11 Taylor’s Theorem 291 and by the Squeeze Theorem (11.1.3) lim N→∞ xN+1 =0 (N + 1)! as desired. Essentially the same argument works for cos x and ex ; unfortunately, it is more difficult to show that most functions are equal to their Maclaurin series. EXAMPLE 11.11.3 Find a polynomial approximation for ex near x = 2 accurate to ±0.005. From Taylor’s theorem: N X ez e2 (x − 2)n + (x − 2)N+1 , e = n! (N + 1)! n=0 x since f (n) (x) = ex for all n. We are interested in x near 2, and we need to keep |(x−2)N+1 | in check, so we may as well specify that |x − 2| ≤ 1, so x ∈ [1, 3]. Also ez e3 ≤ , (N + 1)! (N + 1)! so we need to find an N that makes e3 /(N + 1)! ≤ 0.005. This time N = 5 makes e3 /(N + 1)! < 0.0015, so the approximating polynomial is ex = e2 + e2 (x − 2) + e2 e2 e2 e2 (x − 2)2 + (x − 2)3 + (x − 2)4 + (x − 2)5 ± 0.0015. 2 6 24 120 This presents an additional problem for approximation, since we also need to approximate e2 , and any approximation we use will increase the error, but we will not pursue this complication. Note well that in these examples we found polynomials of a certain accuracy only on a small interval, even though the series for sin x and ex converge for all x; this is typical. To get the same accuracy on a larger interval would require more terms. Exercises 11.11. 1. Find a polynomial approximation for cos x on [0, π], accurate to ±10−3 ⇒ 2. How many terms of the series for ln x centered at 1 are required so that the guaranteed error on [1/2, 3/2] is at most 10−3 ? What if the interval is instead [1, 3/2]? ⇒ 3. Find the first three nonzero terms in the Taylor series for tan x on [−π/4, π/4], and compute the guaranteed error term as given by Taylor’s theorem. (You may want to use Sage or a similar aid.) ⇒ 292 Chapter 11 Sequences and Series 4. Show that cos x is equal to its Taylor series for all x by showing that the limit of the error term is zero as N approaches infinity. 5. Show that ex is equal to its Taylor series for all x by showing that the limit of the error term is zero as N approaches infinity. 11.12 Additional exer ises These problems require the techniques of this chapter, and are in no particular order. Some problems may be done in more than one way. Determine whether the series converges. ∞ X n 1. ⇒ 2 n +4 n=0 1 1 1 1 + + + + ··· ⇒ 1·2 3·4 5·6 7·8 ∞ X n ⇒ 3. 2 + 4)2 (n n=0 2. ∞ X n! 4. ⇒ n 8 n=0 5 7 9 3 + − + + ··· ⇒ 4 8 12 16 ∞ X 1 √ ⇒ 6. n2 + 4 n=0 5. 1 − 7. ∞ X sin3 (n) ⇒ n2 n=0 8. ∞ X n ⇒ en n=0 9. ∞ X 10. n=0 ∞ X n=1 n! ⇒ 1 · 3 · 5 · · · (2n − 1) 1 √ ⇒ n n 1 2 3 4 + + + + ··· ⇒ 2·3·4 3·4·5 4·5·6 5·6·7 ∞ X 1 · 3 · 5 · · · (2n − 1) ⇒ 12. (2n)! n=1 11. ∞ X 6n 13. ⇒ n! n=0 14. ∞ X (−1)n−1 √ ⇒ n n=1 11.12 15. 293 ∞ X 2n 3n−1 ⇒ n! n=1 16. 1 + 17. Additional exercises ∞ X n=1 52 54 56 58 + + + + ··· ⇒ 22 (2 · 4)2 (2 · 4 · 6)2 (2 · 4 · 6 · 8)2 sin(1/n) ⇒ Find the interval and radius of convergence; you need not check the endpoints of the intervals. ∞ X 2n n 18. x ⇒ n! n=0 19. 20. ∞ X n=0 ∞ X n=1 xn ⇒ 1 + 3n xn ⇒ n3n 1 x3 1 · 3 x5 1 · 3 · 5 x7 + + + ··· ⇒ 2 3 2·4 5 2·4·6 7 ∞ X n! n x ⇒ 22. 2 n n=1 21. x + 23. ∞ X (−1)n 2n x ⇒ n 2 3n n=1 24. ∞ X (x − 1)n ⇒ n! n=0 Find a series for each function, using the formula for Maclaurin series and algebraic manipulation as appropriate. 25. 2x ⇒ 26. ln(1 + x) ⇒   1+x 27. ln ⇒ 1−x √ 28. 1 + x ⇒ 1 ⇒ 29. 1 + x2 30. arctan(x) ⇒ 31. Use the answer to the previous problem to discover a series for a well-known mathematical constant. ⇒