Calculus3Sequences and Series5 of 6
Sequences and Series · 5

Alternating Series and Absolute Convergence

1

When the signs alternate

If we take the harmonic series 1+12+13+14+⋯\rgt{1} + \rgt{\frac12} + \rgt{\frac13} + \rgt{\frac14} + \cdots and switch every second sign to a minus, does the new series add up to a number? The harmonic series diverges to infinity, and none of our tests decide the new one. The terms 1n\frac1n approach 00, so the nth-term test tells us nothing, and the integral test and the comparison tests only work when every term is positive. The factor (−1)n+1(-1)^{n+1} is 1\rgt{1} for odd nn and −1\lft{-1} for even nn, so we can write the new series as ∑n=1∞(−1)n+1n\sum_{n=1}^\infty \frac{(-1)^{n+1}}{n}.

All positive terms
1+12+13+14+⋯\displaystyle {\rgt{1} + \rgt{\frac12} + \rgt{\frac13} + \rgt{\frac14} + \cdots}
Alternating signs
1−12+13−14+⋯\displaystyle {\rgt{1} - \lft{\frac12} + \rgt{\frac13} - \lft{\frac14} + \cdots} =∑n=1∞(−1)n+1n\displaystyle = \sum_{n=1}^{\infty} \frac{(-1)^{n+1}}{n}

Let's track the partial sums as positions on a number line, starting at 00. A positive term moves us to the right by its size, and a negative term moves us to the left. In the harmonic series every step goes right, so the positions go to infinity. Here the steps switch direction.

00
11
12\dfrac{1}{2}
4760\dfrac{47}{60}
712\dfrac{7}{12}
56\dfrac{5}{6}
0+1=11−12=1212+13=5656−14=712712+15=4760\displaystyle \begin{aligned} 0 + \rgt{1} &= 1\\[6pt] 1 - \lft{\frac12} &= \frac12\\[6pt] \frac12 + \rgt{\frac13} &= \frac56\\[6pt] \frac56 - \lft{\frac14} &= \frac{7}{12}\\[6pt] \frac{7}{12} + \rgt{\frac15} &= \frac{47}{60}\end{aligned}
Does 1−12+13−14+⋯\rgt{1} - \lft{\frac12} + \rgt{\frac13} - \lft{\frac14} + \cdots converge?
1−12+13−14+⋯\displaystyle \rgt{1} - \lft{\frac12} + \rgt{\frac13} - \lft{\frac14} + \cdots =  ?\displaystyle = \;?
Text solution
Each step turns back by lessBecause 13<12\rgt{\frac13} < \lft{\frac12}, stepping right from 12\frac12 doesn't reach back to 11. Because 14<13\lft{\frac14} < \rgt{\frac13}, stepping left from 56\frac56 doesn't reach back to 12\frac12. So each new position stays inside the stretch of the step before it.
12<56<1\displaystyle \frac12 < \frac56 < 1
12<712<56\displaystyle \frac12 < \frac{7}{12} < \frac56
The stretches close in on one numberThe stretch lengths are the step sizes 12,13,14,…\strc{\frac12}, \strc{\frac13}, \strc{\frac14}, \dots, which approach 00. So only one number SS is inside every stretch, and the partial sums approach it from both sides.
12<712<⋯<S\displaystyle \frac12 < \frac{7}{12} < \cdots < S
S<⋯<4760<56<1\displaystyle S < \cdots < \frac{47}{60} < \frac56 < 1

So ∑(−1)n+1n\sum \frac{(-1)^{n+1}}{n} converges, even though ∑1n\sum \frac1n, with the very same sizes, diverges. The same reasoning works for any series whose signs alternate. We write ∣an∣\sz{|a_n|} for the size of the nth term, its absolute value. If the sizes decrease, each position stays inside the previous stretch, and if they also approach 00, the series converges. This is the alternating series test.

The alternating series test
Signs alternate
the signs of an alternate\displaystyle \text{the signs of } a_n \text{ alternate}
Sizes decrease
∣an+1∣\displaystyle \sz{|a_{n+1}|} ≤∣an∣\displaystyle \le \sz{|a_n|}
Sizes approach 0
∣an∣→0\displaystyle \sz{|a_n|} \to 0
⟹ ∑an converges\displaystyle \Longrightarrow\ \sum a_n \text{ converges}
2

When both the top and the bottom increase

Let's test an alternating series where both the numerator and the denominator increase. The sizes of its terms are ln⁡nn\sz{\frac{\ln n}{n}}, and since ln⁡n\ln n and nn both grow, we can't tell just by looking whether they decrease. The first few sizes even go up before they come down.

11
22
33
44
55
66
77
88
0.10.1
0.20.2
0.30.3
xx
ee
0.347\htmlClass{lc-violet}{0.347}
0.366\htmlClass{lc-violet}{0.366}
f(x)=ln⁡xx\htmlClass{lc-violet}{f(x) = \dfrac{\ln x}{x}}
ln⁡22≈0.347ln⁡33≈0.366ln⁡44≈0.347ln⁡55≈0.322\displaystyle \begin{aligned} \sz{\frac{\ln 2}{2}} &\approx 0.347\\[6pt] \sz{\frac{\ln 3}{3}} &\approx 0.366\\[6pt] \sz{\frac{\ln 4}{4}} &\approx 0.347\\[6pt] \sz{\frac{\ln 5}{5}} &\approx 0.322\end{aligned}
Does ∑n=2∞(−1)nln⁡nn\sum_{n=2}^{\infty} \frac{(-1)^n \ln n}{n} converge or diverge?
∑n=2∞(−1)nln⁡nn\displaystyle \sum_{n=2}^{\infty} \frac{(-1)^n \ln n}{n}
converges or diverges?\displaystyle \text{converges or diverges?}
Text solution
Check the limit firstln⁡x\ln x and xx both go to infinity, so we use L'Hôpital's rule. The sizes approach 00.
lim⁡x→∞ln⁡xx\displaystyle \lim_{x\to\infty} \sz{\frac{\ln x}{x}} =lim⁡x→∞ 1x 1\displaystyle = \lim_{x\to\infty} \frac{\,\frac1x\,}{1} =0\displaystyle = 0
Find where the sizes decreaseThe derivative of f(x)=ln⁡xx\sz{f(x) = \frac{\ln x}{x}} is negative when ln⁡x>1\ln x > 1, that is x>e≈2.72x > e \approx 2.72. So the sizes decrease from n=3n = 3 on.
f′(x)\displaystyle f'(x) =1−ln⁡xx2<0  for x>e\displaystyle = \frac{1 - \ln x}{x^2} < 0 \ \text{ for } x > e
Split off the first termFrom n=3n = 3 on, the sizes decrease to 00, so that part converges. Adding ln⁡22\rgt{\frac{\ln 2}{2}} only shifts the sum by a fixed number, so the whole series converges.
∑n=2∞(−1)nln⁡nn=ln⁡22+∑n=3∞(−1)nln⁡nn\displaystyle \begin{aligned} &\sum_{n=2}^{\infty} \frac{(-1)^n \ln n}{n}\\ &\quad = \rgt{\frac{\ln 2}{2}} + \sum_{n=3}^{\infty} \frac{(-1)^n \ln n}{n}\end{aligned}

Checking the limit first matters. Take 2−32+43−54+⋯\rgt{2} - \lft{\frac32} + \rgt{\frac43} - \lft{\frac54} + \cdots. Its sizes decrease, but they approach 11 instead of 00, so it diverges by the nth-term test.

∑n=1∞(−1)n+1(n+1)n\displaystyle \sum_{n=1}^{\infty} \frac{(-1)^{n+1}(n+1)}{n} =2−32+43−54+⋯  ?\displaystyle = {\rgt{2} - \lft{\frac32} + \rgt{\frac43} - \lft{\frac54} + \cdots}\;?
∣an∣\displaystyle \sz{|a_n|} =n+1n→1≠0\displaystyle = {\sz{\frac{n+1}{n}} \to 1 \ne 0}

The alternating series test only tells us when a series converges. If the sizes approach 00 but aren't decreasing from any point on, it gives no conclusion, and we have to try another test.

You'll also see alternating series written with cos⁡(nπ)\cos(n\pi) instead of (−1)n(-1)^n. They're the same thing, so a series ∑cos⁡(nπ)bn\sum \cos(n\pi)\sz{b_n} with bn>0\sz{b_n} > 0 alternates, and we run the same two checks on bn\sz{b_n}.

cos⁡π=−1,cos⁡2π=1,cos⁡3π=−1\displaystyle {\cos \pi = \lft{-1}, \qquad \cos 2\pi = \rgt{1}, \qquad \cos 3\pi = \lft{-1}}
cos⁡(nπ)\displaystyle \cos(n\pi) =(−1)n\displaystyle = (-1)^n
cos⁡π=−1cos⁡2π=1cos⁡3π=−1cos⁡(nπ)=(−1)n\displaystyle \begin{aligned} \cos \pi &= \lft{-1}\\ \cos 2\pi &= \rgt{1}\\ \cos 3\pi &= \lft{-1}\\[4pt] \cos(n\pi) &= (-1)^n\end{aligned}
3

Estimating the sum of an alternating series

If we stop after four terms of 1−12+13−14+⋯\rgt{1} - \lft{\frac12} + \rgt{\frac13} - \lft{\frac14} + \cdots, we get 712\frac{7}{12}. How close is 712\frac{7}{12} to the actual sum SS, and is it too high or too low? The number line from the first section answers both.

Find the error bound and whether the estimate is too high or too low after 4 terms and after 5 terms of 1−12+13−⋯\rgt{1} - \lft{\frac12} + \rgt{\frac13} - \cdots
00
11
12\dfrac{1}{2}
712\dfrac{7}{12}
56\dfrac{5}{6}
Text solution
Look at the first term left outAfter four terms, that's +15\rgt{+\frac15}, a step from 712\frac{7}{12} to 4760\frac{47}{60}. Every later sum stays inside that stretch, so SS does too.
712<S<712+15\displaystyle {\frac{7}{12} < S < \frac{7}{12} + \rgt{\frac15}} =4760\displaystyle = \frac{47}{60}
Read off the error and the sideThe stretch is only 15\rgt{\frac15} long, so the error is less than 15\rgt{\frac15}. The step went right, so 712\frac{7}{12} is too low.
After five termsThe first term left out is −16\lft{-\frac16}, a step left to 3760\frac{37}{60}. The error is less than 16\lft{\frac16}, and this time the estimate 4760\frac{47}{60} is too high.
3760\displaystyle \frac{37}{60} =4760−16<S<4760\displaystyle = {\frac{47}{60} - \lft{\frac16} < S < \frac{47}{60}}

This works for any alternating series once the sizes of the terms we leave out decrease toward 00. The sum SS is always between our estimate and that estimate plus the next term.

error\displaystyle \err{\text{error}} ≤∣first term left out∣\displaystyle \le \sz{|\text{first term left out}|}
first term left out>0→estimate too lowfirst term left out<0→estimate too high\displaystyle \begin{aligned} \rgt{\text{first term left out} > 0} &\to \text{estimate too low}\\ \lft{\text{first term left out} < 0} &\to \text{estimate too high}\end{aligned}
first term left out>0→estimate too lowfirst term left out<0→estimate too high\displaystyle \begin{aligned} &\rgt{\text{first term left out} > 0}\\ &\qquad \to \text{estimate too low}\\[6pt] &\lft{\text{first term left out} < 0}\\ &\qquad \to \text{estimate too high}\end{aligned}

Does the error bound come from the last term we added or the first term we left out?

We can also turn the bound around and decide how many terms to add for the accuracy we want.

Find how many terms of ∑n=1∞(−1)n+1n2\sum_{n=1}^{\infty} \frac{(-1)^{n+1}}{n^2} give an error less than 0.005\err{0.005}
∑n=1∞(−1)n+1n2\displaystyle \sum_{n=1}^{\infty} \frac{(-1)^{n+1}}{n^2} =1−14+19−⋯\displaystyle = {\rgt{1} - \lft{\frac14} + \rgt{\frac19} - \cdots}
error<0.005\displaystyle \err{\text{error}} < \err{0.005}
number of terms\displaystyle \text{number of terms} =  ?\displaystyle = \;?
Text solution
Make the first term left out small enoughEach term has size 1n2\sz{\frac{1}{n^2}}, and 0.005\err{0.005} is 1200\err{\frac{1}{200}}, so we need n2>200n^2 > 200.
1n2<1200 → n2>200\displaystyle \sz{\frac{1}{n^2}} < \err{\frac{1}{200}} \ \to\ n^2 > 200
Find the first such term142=19614^2 = 196 isn't enough, but 152=22515^2 = 225 is. So the 15th term is the first one we leave out, and we add the first 14 terms.
1196≈0.0051>0.005\displaystyle \sz{\frac{1}{196}} \approx 0.0051 > \err{0.005}
1225≈0.0044<0.005\displaystyle \sz{\frac{1}{225}} \approx 0.0044 < \err{0.005}
Too high or too lowWhen n=15n = 15, (−1)16=1(-1)^{16} = \rgt{1}, so the 15th term is +1225\rgt{+\frac{1}{225}}, a step right toward SS. The sum of the first 14 terms is slightly too low.
4

When the signs follow no pattern

So far, our series have either had all positive terms or terms that alternate with each step. Now let's look at ∑sin⁡nn2\sum \frac{\sin n}{n^2}, whose signs follow no fixed pattern. The sign of each term comes from sin⁡n\sin n, the y-value of y=sin⁡xy = \sin x at x=nx = n. Since π≈3.14\pi \approx 3.14 and 2π≈6.282\pi \approx 6.28, the signs come in runs.

11
−1-1
xx
2π2\pi
11
22
33
44
55
66
77
y=sin⁡xy = \sin x
sin⁡1, sin⁡2, sin⁡3>0sin⁡4, sin⁡5, sin⁡6<0sin⁡7>0\displaystyle \begin{aligned} \rgt{\sin 1,\ \sin 2,\ \sin 3} &> 0\\[4pt] \lft{\sin 4,\ \sin 5,\ \sin 6} &< 0\\[4pt] \rgt{\sin 7} &> 0\end{aligned}

Because the period 2π2\pi isn't a whole number, the signs never settle into a plus-minus pattern, so the alternating series test doesn't apply. And because some terms are negative, we can't use direct comparison or the integral test either.

Does ∑n=1∞sin⁡nn2\sum_{n=1}^{\infty} \frac{\sin n}{n^2} converge or diverge?
∑n=1∞sin⁡nn2\displaystyle \sum_{n=1}^{\infty} \frac{\sin n}{n^2} =sin⁡1+sin⁡24+sin⁡39+⋯\displaystyle = \sin 1 + \frac{\sin 2}{4} + \frac{\sin 3}{9} + \cdots
converges or diverges?\displaystyle \text{converges or diverges?}
Text solution
Make every term positiveWith absolute values, every term is positive, so we can use direct comparison. Since ∣sin⁡n∣≤1|\sin n| \le 1, we compare with ∑1n2\sum \frac{1}{n^2}, a pp-series with p=2>1p = 2 > 1.
∣sin⁡n∣n2≤1n2∑1n2 converges→ ∑∣sin⁡n∣n2 converges\displaystyle \begin{aligned} &\sz{\frac{|\sin n|}{n^2}} \le \frac{1}{n^2}\\[6pt] &\sum \frac{1}{n^2} \text{ converges}\\ &\qquad \to\ \sz{\sum \frac{|\sin n|}{n^2}} \text{ converges}\end{aligned}
Why the original series convergesThe steps to the right are some of the absolute values, so their total never passes ∑∣an∣\sz{\sum |a_n|} and approaches a finite number. The same goes for the steps to the left. Our position is the difference of the two totals, so it approaches a finite number too.
position\displaystyle \text{position} =total to the right−total to the left\displaystyle = \rgt{\text{total to the right}} - \lft{\text{total to the left}}

That argument works for any series. When the series of absolute values converges, the original series converges too, no matter what signs the terms have, and we say it converges absolutely. When a series converges but its series of absolute values diverges, we say it converges conditionally.

Converges absolutely
∑∣an∣ converges⟹ ∑an converges\displaystyle \begin{aligned} &\sz{\sum |a_n|} \text{ converges}\\ &\quad \Longrightarrow\ \sum a_n \text{ converges}\end{aligned}
Converges conditionally
∑an converges,∑∣an∣ diverges\displaystyle \begin{aligned} &\sum a_n \text{ converges},\\ &\sz{\sum |a_n|} \text{ diverges}\end{aligned}

The alternating harmonic series ∑(−1)n+1n\sum \frac{(-1)^{n+1}}{n} converges conditionally. Its absolute values make the harmonic series, and both its steps to the left and its steps to the right add up to infinity. It still converges, because its signs alternate and its sizes decrease to 00.

12+14+16+⋯\displaystyle \lft{\frac12 + \frac14 + \frac16 + \cdots} =12(1+12+13+⋯)\displaystyle = \lft{\frac12\Big(1 + \frac12 + \frac13 + \cdots\Big)}
1>12,13>14,15>16, …\displaystyle \rgt{1} > \lft{\frac12}, \quad \rgt{\frac13} > \lft{\frac14}, \quad \rgt{\frac15} > \lft{\frac16}, \ \ldots

An alternating sign pattern doesn't automatically mean conditional convergence. For ∑(−1)n+1n2\sum \frac{(-1)^{n+1}}{n^2}, the absolute values give ∑1n2\sz{\sum \frac{1}{n^2}}, which converges, so it converges absolutely.

∑(−1)n+1n2 converges∑1n2 converges\displaystyle \begin{aligned} &\sum \frac{(-1)^{n+1}}{n^2} \text{ converges}\\ &\sz{\sum \frac{1}{n^2}} \text{ converges}\end{aligned}
absolutely\displaystyle \text{absolutely}
∑(−1)n+1n converges∑1n diverges\displaystyle \begin{aligned} &\sum \frac{(-1)^{n+1}}{n} \text{ converges}\\ &\sz{\sum \frac1n} \text{ diverges}\end{aligned}
conditionally\displaystyle \text{conditionally}

What does removing the signs tell us about convergence?

5

Absolutely, conditionally, or diverges

When we have an alternating series, we want to know whether it converges absolutely, converges conditionally, or diverges. Testing in the right order saves work.

First check
an→0 ?\displaystyle a_n \to 0\,?
no→diverges\displaystyle \text{no} \to \text{diverges}
Second check
∑∣an∣ converges?\displaystyle \sz{\sum |a_n|} \text{ converges?}
yes→absolutely\displaystyle \text{yes} \to \text{absolutely}
Third check
alternating seriestest passes?\displaystyle \begin{gathered}\text{alternating series}\\ \text{test passes?}\end{gathered}
yes→conditionally\displaystyle \text{yes} \to \text{conditionally}

If we only run the alternating series test on ∑(−1)nn2\sum \frac{(-1)^n}{n^2}, we might say it converges conditionally, but ∑1n2\sz{\sum \frac{1}{n^2}} converges, so it converges absolutely. The other mistake is stopping after the series of absolute values diverges. For ∑(−1)nn\sum \frac{(-1)^n}{\sqrt{n}}, the series ∑1n\sz{\sum \frac{1}{\sqrt{n}}} diverges, but 1n\sz{\frac{1}{\sqrt{n}}} decreases to 00, so it converges conditionally.

Does ∑n=1∞(−1)nnn2+1\sum_{n=1}^\infty \frac{(-1)^n n}{n^2+1} converge absolutely, converge conditionally, or diverge?
∑n=1∞(−1)n nn2+1\displaystyle \sum_{n=1}^{\infty} \frac{(-1)^n\, n}{n^2+1}
absolutely, conditionally,\displaystyle \text{absolutely, conditionally,}
or diverges?\displaystyle \text{or diverges?}
Text solution
The terms approach 0The denominator has degree 22 and the numerator has degree 11, so the nth-term test doesn't decide it.
∣an∣=nn2+1→0\displaystyle \sz{|a_n| = \frac{n}{n^2+1}} \to 0
The absolute values divergeFor large nn, nn2+1\sz{\frac{n}{n^2+1}} behaves like nn2=1n\frac{n}{n^2} = \frac1n. A limit comparison with 1n\frac1n gives 11, so ∑∣an∣\sz{\sum |a_n|} diverges like the harmonic series, and the series doesn't converge absolutely.
  nn2+1  1n=n2n2+1→1∑1n diverges → ∑∣an∣ diverges\displaystyle \begin{aligned} &\frac{\;\sz{\dfrac{n}{n^2+1}}\;}{\dfrac1n} = \frac{n^2}{n^2+1} \to 1\\[6pt] &\sum \frac1n \text{ diverges} \ \to\ \sz{\sum |a_n|} \text{ diverges}\end{aligned}
The sizes decreaseThe top and the bottom both increase, so we take the derivative of f(x)=xx2+1\sz{f(x) = \frac{x}{x^2+1}}. Its numerator 1−x21 - x^2 is negative for x>1x > 1, so the alternating series test passes. The series converges, but not absolutely, so it converges conditionally.
f′(x)=(x2+1)−2x2(x2+1)2=1−x2(x2+1)2<0  for x>1\displaystyle \begin{aligned} f'(x) &= \frac{(x^2+1) - 2x^2}{(x^2+1)^2}\\ &= \frac{1 - x^2}{(x^2+1)^2} < 0 \ \text{ for } x > 1\end{aligned}

The same three checks sort the whole alternating p-series by the value of pp.

The alternating p-series
∑(−1)n+1np\displaystyle \sum \frac{(-1)^{n+1}}{n^p}
p>1→ absolutely0<p≤1→ conditionallyp≤0→ ∣an∣=n−p≥1, diverges\displaystyle \begin{aligned} p &> 1 &&\to\ \text{absolutely}\\ 0 < p &\le 1 &&\to\ \text{conditionally}\\ p &\le 0 &&\to\ \sz{|a_n| = n^{-p}} \ge 1, \ \text{diverges}\end{aligned}
p>1→ absolutely0<p≤1→ conditionallyp≤0→ ∣an∣=n−p≥1,→  diverges\displaystyle \begin{aligned} p &> 1 &&\to\ \text{absolutely}\\ 0 < p &\le 1 &&\to\ \text{conditionally}\\ p &\le 0 &&\to\ \sz{|a_n| = n^{-p}} \ge 1,\\ &&&\phantom{{}\to{}}\ \ \text{diverges}\end{aligned}

Summary

The alternating series test
00
11
SS
signs alternate∣an∣ decreasingfrom some n on∣an∣→0⟹ ∑an converges\displaystyle \begin{aligned} &\text{signs alternate}\\ &\sz{|a_n|} \text{ decreasing}\\ &\quad \text{from some } n \text{ on}\\ &\sz{|a_n|} \to 0\\[4pt] &\Longrightarrow\ \sum a_n \text{ converges}\end{aligned}

If the terms alternate in sign, their sizes ∣an∣\sz{|a_n|} decrease, and their sizes approach 00, then the series converges. On the number line, each position stays inside the stretch of the step before it, and the partial sums approach a single value SS.

cos⁡(nπ)\cos(n\pi) equals (−1)n(-1)^n, so it alternates in the exact same way. If the top and bottom of ∣an∣\sz{|a_n|} both increase, write it as a function f(x)\sz{f(x)} with xx in place of nn, like ln⁡xx\sz{\frac{\ln x}{x}}. If the derivative f′(x)f'(x) is negative for large xx, the function is decreasing, so the sizes decrease from that point on. The sizes only need to decrease from some term on, because the first few terms never decide whether a series converges. If ∣an∣\sz{|a_n|} doesn't approach 00, the series diverges by the nth-term test. The alternating series test only ever shows convergence.

Alternating series error bound
error\displaystyle \err{\text{error}} ≤∣first term left out∣\displaystyle \le \sz{|\text{first term left out}|}
1−12+13−14=712first term left out +15→ within 15, too low\displaystyle \begin{aligned} &\rgt{1} - \lft{\frac12} + \rgt{\frac13} - \lft{\frac14} = \frac{7}{12}\\ &\text{first term left out } \rgt{+\frac15}\\ &\quad \to\ \text{within } \rgt{\frac15}, \text{ too low}\end{aligned}

Once the omitted term sizes decrease toward 0, the next step on the number line is the first term left out. The total sum SS is between where you stopped and that next position. So the error is at most the size of the first term left out.

If the first term left out is positive, the next step moves right, so the estimate is too low. If it's negative, the next step moves left, so the estimate is too high. To find how many terms are needed for an error less than a given bound, find the position of the first term whose size is less than that bound, and add all the terms before it.

Absolute and conditional convergence
∑∣an∣ converges ⟹ ∑an converges (absolutely)\displaystyle \sz{\sum |a_n|} \text{ converges} \ \Longrightarrow\ \sum a_n \text{ converges (absolutely)}
∑∣an∣ converges⟹ ∑an converges(absolutely)\displaystyle \begin{aligned} &\sz{\sum |a_n|} \text{ converges}\\ &\quad \Longrightarrow\ \sum a_n \text{ converges}\\ &\qquad \text{(absolutely)}\end{aligned}
∑an converges, ∑∣an∣ diverges: conditionally\displaystyle \sum a_n \text{ converges}, \ \sz{\sum |a_n|} \text{ diverges}: \ \text{conditionally}
∑an converges,∑∣an∣ diverges:conditionally\displaystyle \begin{aligned} &\sum a_n \text{ converges},\\ &\sz{\sum |a_n|} \text{ diverges}:\\ &\qquad \text{conditionally}\end{aligned}

If the series of absolute values ∑∣an∣\sz{\sum |a_n|} converges, then the original series ∑an\sum a_n converges too, no matter what signs the terms have. We say it converges absolutely. For series like ∑sin⁡nn2\sum \frac{\sin n}{n^2}, where the signs follow no regular pattern, use ∣sin⁡n∣≤1|\sin n| \le 1 to compare ∑∣an∣\sz{\sum |a_n|} with ∑1n2\sum \frac{1}{n^2}.

If ∑an\sum a_n converges but ∑∣an∣\sz{\sum |a_n|} diverges, the series converges conditionally. The alternating harmonic series 1−12+13−14+⋯\rgt{1} - \lft{\frac12} + \rgt{\frac13} - \lft{\frac14} + \cdots converges conditionally because it passes the alternating series test, even though ∑1n\sz{\sum \frac1n} diverges.

The order of checks
∑(−1)nnp\displaystyle \sum \frac{(-1)^n}{n^p}
p>1→ absolutely0<p≤1→ conditionallyp≤0→ diverges\displaystyle \begin{aligned} p &> 1 &&\to\ \text{absolutely}\\ 0 < p &\le 1 &&\to\ \text{conditionally}\\ p &\le 0 &&\to\ \text{diverges}\end{aligned}

To test an alternating series, first check if the terms approach 00. If not, it diverges by the nth-term test. If they do approach 00, test the series of absolute values ∑∣an∣\sz{\sum |a_n|}. If ∑∣an∣\sz{\sum |a_n|} converges, the series converges absolutely, and you never have to check whether the sizes decrease. If ∑∣an∣\sz{\sum |a_n|} diverges, use the alternating series test. If the sizes decrease to 00, it converges conditionally.

Take the alternating p-series ∑(−1)nnp\sum \frac{(-1)^n}{n^p}. When p>1p > 1, ∑1np\sz{\sum \frac{1}{n^p}} converges, so the series converges absolutely. When 0<p≤10 < p \le 1, ∑1np\sz{\sum \frac{1}{n^p}} diverges, but the sizes decrease to 00, so the alternating series converges conditionally. When p≤0p \le 0, the terms don't approach 00, so the series diverges.

6

Practice

(−1)n(-1)^n and (−1)n+1(-1)^{n+1} differ only in which sign comes first. Either way, the signs alternate.

1
Does ∑n=1∞(−1)n+1n\displaystyle\sum_{n=1}^{\infty} \frac{(-1)^{n+1}}{\sqrt{n}} converge absolutely, converge conditionally, or diverge?

Look at the series of absolute values, ∑1n\sum \frac{1}{\sqrt n}. Which p-series is that?

The series of absolute values diverges, but the signs alternate. Does 1n\frac{1}{\sqrt n} decrease and approach 00?

If you got diverges The series of absolute values diverges, but the signs alternate, so check the alternating series test before deciding.

As nn goes to infinity, the denominator increases, so the size of each term approaches 00.
∣an∣=1n→0\displaystyle \sz{|a_n| = \frac{1}{\sqrt n}} \to 0
Test for absolute convergenceThe series of absolute values is a p-series with p=12p = \frac12, which is less than 11, so it diverges.
∑1n=∑1n12,p\displaystyle \sz{\sum \frac{1}{\sqrt n} = \sum \frac{1}{n^{\frac12}}}, \quad p =12\displaystyle = \frac12 ≤1 → ∑∣an∣ diverges\displaystyle \le 1 \ \to\ \sz{\sum |a_n|} \text{ diverges}
Test for conditional convergenceThe denominator n\sqrt n increases while the numerator stays 11, so the sizes decrease to 00, and the alternating series test passes.
1n+1<1n,1n→0 → converges conditionally\displaystyle \sz{\frac{1}{\sqrt{n+1}}} < \sz{\frac{1}{\sqrt n}}, \quad \sz{\frac{1}{\sqrt n}} \to 0 \ \to\ \text{converges conditionally}
2
Does ∑n=1∞(−1)n n2n+1\displaystyle\sum_{n=1}^{\infty} \frac{(-1)^n\, n}{2n+1} converge absolutely, converge conditionally, or diverge?

What does n2n+1\frac{n}{2n+1} approach as nn goes to infinity?

If you got converges conditionally Check the limit of the terms as n→∞n \to \infty. A series can converge only if its terms approach 00.

The terms alternate, but their sizes approach 12\frac12 rather than 00, so by the nth-term test the series diverges.
∣an∣=n2n+1:13, 25, 37, …\displaystyle \sz{|a_n| = \frac{n}{2n+1}}: \quad \sz{\frac13},\ \sz{\frac25},\ \sz{\frac37},\ \ldots
lim⁡n→∞n2n+1\displaystyle \lim_{n\to\infty} \sz{\frac{n}{2n+1}} =lim⁡n→∞12+1n\displaystyle = \lim_{n\to\infty} \frac{1}{2 + \frac1n} =12≠0 → diverges\displaystyle = \frac12 \ne 0 \ \to\ \text{diverges}
3
Does ∑n=1∞cos⁡nn2+1\displaystyle\sum_{n=1}^{\infty} \frac{\cos n}{n^2+1} converge absolutely, converge conditionally, or diverge?

In radians, what are the signs of cos⁡1\cos 1, cos⁡2\cos 2, and cos⁡3\cos 3?

Since ∣cos⁡n∣≤1|\cos n| \le 1, what is an upper bound for the size of each term?

If you got diverges The signs do not alternate at every step, but that only means the alternating series test cannot be used. Test the series of absolute values directly.

The signs do not alternate each step, so the alternating series test does not apply.
cos⁡1>0,cos⁡2<0,cos⁡3<0,cos⁡4<0,cos⁡5>0\displaystyle \rgt{\cos 1} > 0, \quad \lft{\cos 2} < 0, \quad \lft{\cos 3} < 0, \quad \lft{\cos 4} < 0, \quad \rgt{\cos 5} > 0
Test for absolute convergenceSince ∣cos⁡n∣≤1|\cos n| \le 1, we can compare the absolute values directly to 1n2\frac{1}{n^2}.
∣cos⁡n∣n2+1\displaystyle \sz{\frac{|\cos n|}{n^2+1}} ≤1n2+1<1n2,∑1n2 converges → converges absolutely\displaystyle \le \frac{1}{n^2+1} < \frac{1}{n^2}, \qquad \sum \frac{1}{n^2} \text{ converges} \ \to\ \text{converges absolutely}
4
Does ∑n=2∞(−1)nln⁡nn\displaystyle\sum_{n=2}^{\infty} \frac{(-1)^n \ln n}{\sqrt{n}} converge absolutely, converge conditionally, or diverge?

For n≥3n \ge 3, is ln⁡nn\frac{\ln n}{\sqrt n} greater than or less than 1n\frac{1}{\sqrt n}?

Take the derivative of f(x)=ln⁡xxf(x) = \frac{\ln x}{\sqrt x}. For which values of xx is f′(x)<0f'(x) < 0?

If you got diverges The sizes increase at first, but the alternating series test only requires them to decrease from some term on.

Check the limit of the termsBoth the numerator and denominator go to infinity, so L'Hôpital's rule shows that the term sizes approach 00.
lim⁡x→∞ln⁡xx\displaystyle \lim_{x\to\infty} \sz{\frac{\ln x}{\sqrt x}} =lim⁡x→∞1x12x\displaystyle = \lim_{x\to\infty} \frac{\frac{1}{x}}{\frac{1}{2\sqrt x}} =lim⁡x→∞2x\displaystyle = \lim_{x\to\infty} \frac{2}{\sqrt x} =0\displaystyle = 0
Test for absolute convergenceFor n≥3n \ge 3, ln⁡n>1\ln n > 1, so compare the absolute values with 1n\frac{1}{\sqrt n}.
n\displaystyle n ≥3: ln⁡n>1 → ln⁡nn>1n,∑1n diverges → ∑∣an∣ diverges\displaystyle \ge 3: \ \ln n > 1 \ \to\ \sz{\frac{\ln n}{\sqrt n}} > \frac{1}{\sqrt n}, \qquad \sum \frac{1}{\sqrt n} \text{ diverges} \ \to\ \sz{\sum |a_n|} \text{ diverges}
Find where the sizes decreaseThe sizes increase at first, so use the derivative of f(x)=ln⁡xx\sz{f(x) = \frac{\ln x}{\sqrt x}} to find where they begin decreasing.
f(x)=ln⁡xx,f′(x)\displaystyle \sz{f(x) = \frac{\ln x}{\sqrt x}}, \qquad f'(x) =1xx−ln⁡x⋅12xx\displaystyle = \frac{\frac{1}{x}\sqrt x - \ln x \cdot \frac{1}{2\sqrt x}}{x} =2−ln⁡x2x32<0  for x>e2≈7.39\displaystyle = \frac{2 - \ln x}{2x^{\frac32}} < 0 \ \text{ for } x > e^2 \approx 7.39
ln⁡77≈0.7355,ln⁡88≈0.7352,ln⁡99≈0.7324\displaystyle \sz{\frac{\ln 7}{\sqrt 7}} \approx 0.7355, \qquad \sz{\frac{\ln 8}{\sqrt 8}} \approx 0.7352, \qquad \sz{\frac{\ln 9}{\sqrt 9}} \approx 0.7324
The sizes decrease from n=8n = 8 on and approach 00, so the alternating series test passes.
decreasing from n\displaystyle \text{decreasing from } n =8 on,∣an∣→0 → converges conditionally\displaystyle = 8 \text{ on}, \quad \sz{|a_n|} \to 0 \ \to\ \text{converges conditionally}
5
The sum of the first three terms of ∑n=1∞(−1)n+1n3\displaystyle\sum_{n=1}^{\infty} \frac{(-1)^{n+1}}{n^3} is used to approximate its sum SS. (a) Give an upper bound for the error. (b) Is the approximation greater than SS or less than SS?

Which term is the first one left out of the three-term sum?

(b)\text{(b)}

Is the first term left out positive or negative?

Add the first three terms to find the estimate.
1−18+127\displaystyle \rgt{1} - \lft{\frac18} + \rgt{\frac{1}{27}} =216−27+8216\displaystyle = \frac{216 - 27 + 8}{216} =197216\displaystyle = \frac{197}{216}
Find the error boundThe error is at most the absolute value of the first term left out, which is the term for n=4n = 4.
first term left out: −143=−164 → error\displaystyle \text{first term left out: } \lft{-\frac{1}{4^3} = -\frac{1}{64}} \ \to\ \err{\text{error}} ≤164\displaystyle \le \lft{\frac{1}{64}}
Compare the estimate to SSThe first term left out is negative, so adding the next term would decrease the total. That means the three-term estimate is greater than SS.
−164<0 → 197216>S\displaystyle \lft{-\frac{1}{64}} < 0 \ \to\ \frac{197}{216} > S
6
For which values of pp does ∑n=1∞(−1)nnp\displaystyle\sum_{n=1}^{\infty} \frac{(-1)^n}{n^p} converge conditionally?

What happens at p=1p = 1, where the series is ∑(−1)nn\sum \frac{(-1)^n}{n}?

If p=0p = 0, what are the terms of the series?

If you got 0<p<10 < p < 1 Check p=1p = 1, where ∑1n\sum \frac{1}{n} diverges but the alternating series ∑(−1)nn\sum \frac{(-1)^n}{n} converges.

If you got p>0p > 0 For p>1p > 1, the absolute values form a convergent pp-series, so the series converges absolutely.

If you got p≤1p \le 1 For p≤0p \le 0, the terms do not approach 00, so the series diverges.

For p>1p > 1, the series of absolute values converges.
p>1: ∑1np converges → absolutely\displaystyle p > 1: \ \sz{\sum \frac{1}{n^p}} \text{ converges} \ \to\ \text{absolutely}
For 0<p≤10 < p \le 1, the series of absolute values diverges, but the sizes decrease to 00.
0<p\displaystyle 0 < p ≤1: ∑1np diverges, 1np decreasing, 1np→0 → conditionally\displaystyle \le 1: \ \sz{\sum \frac{1}{n^p}} \text{ diverges}, \ \sz{\frac{1}{n^p}} \text{ decreasing}, \ \sz{\frac{1}{n^p}} \to 0 \ \to\ \text{conditionally}
For p≤0p \le 0, the terms do not approach 00.
p\displaystyle p ≤0: ∣an∣=n−p\displaystyle \le 0: \ \sz{|a_n| = n^{-p}} ≥1 → diverges\displaystyle \ge 1 \ \to\ \text{diverges}