Calculus3Sequences and Series3 of 6
Sequences and Series · 3

Integral Test and p-Series

1

Comparing a series to an integral

We already know that the harmonic series, 1+12+13+⋯\ser{1 + \frac12 + \frac13 + \cdots}, diverges to infinity. But what happens if we square each denominator? The terms 1n2\ser{\frac{1}{n^2}} go to 00, but so do the terms of the harmonic series, so the nth-term test can't tell us whether this total is finite.

Running totals look very different for the two series:

1+12+13+⋯\displaystyle \ser{1 + \frac12 + \frac13 + \cdots}
∑1n\displaystyle \ser{\sum \frac{1}{n}}
1+14+19+⋯\displaystyle \ser{1 + \frac14 + \frac19 + \cdots}
∑1n2\displaystyle \ser{\sum \frac{1}{n^2}}
After 100 terms
≈5.19\displaystyle \approx \ser{5.19}
≈1.635\displaystyle \approx \ser{1.635}
After 1,000 terms
≈7.49\displaystyle \approx \ser{7.49}
≈1.644\displaystyle \approx \ser{1.644}
After 1,000,000 terms
≈14.39\displaystyle \approx \ser{14.39}
≈1.645\displaystyle \approx \ser{1.645}

But the harmonic series is only at about 14.39\ser{14.39} after a million terms, and it still goes to infinity, so numbers alone can't prove anything. Since every term is positive, the running total only increases, and one fixed number it never passes would prove convergence. We'll get that number from an area we already know is finite.

∫1∞dxx2\displaystyle \crv{\int_1^\infty \frac{dx}{x^2}} =lim⁡b→∞(1−1b)\displaystyle = \lim_{b\to\infty}\Big(1 - \frac1b\Big) =1\displaystyle = \crv{1}
∫1∞dxx\displaystyle \crv{\int_1^\infty \frac{dx}{x}} =lim⁡b→∞ln⁡b\displaystyle = \lim_{b\to\infty}\ln b =∞\displaystyle = \crv{\infty}

Each term becomes a rectangle of width 11 and height 1n2\ser{\frac{1}{n^2}}, placed on the interval from n−1n - 1 to nn under the curve y=1x2\crv{y = \frac{1}{x^2}}.

00
11
22
33
44
55
66
xx
yy
1\htmlClass{lc-violet}{1}
14\htmlClass{lc-blue}{\dfrac{1}{4}}
19\htmlClass{lc-blue}{\dfrac{1}{9}}
116\htmlClass{lc-blue}{\dfrac{1}{16}}
125\htmlClass{lc-blue}{\dfrac{1}{25}}
136\htmlClass{lc-blue}{\dfrac{1}{36}}
1\htmlClass{lc-pink}{1}
y=1x2\htmlClass{lc-pink}{y = \dfrac{1}{x^2}}
1x2\displaystyle \crv{\frac{1}{x^2}} ≥1n2for n−1≤x≤n\displaystyle \ge \ser{\frac{1}{n^2}} \quad {\text{for } n - 1 \le x \le n}
14+19+116+⋯≤∫1∞dxx2=1\displaystyle \begin{aligned}\ser{\frac14 + \frac19 + \frac1{16} + \cdots} &\le \crv{\int_1^\infty \frac{dx}{x^2}}\\ &= \crv{1}\end{aligned}
Does 1+14+19+⋯\ser{1 + \frac14 + \frac19 + \cdots} converge?
1+14+19+⋯\displaystyle \ser{1 + \frac14 + \frac19 + \cdots}
converges?\displaystyle \text{converges?}
Text solution
Fit the rectangles under the curveBecause 1x2\crv{\frac{1}{x^2}} is decreasing, on [n−1,n][n - 1, n] it's lowest at x=nx = n, where its height is 1n2\ser{\frac{1}{n^2}}. So every rectangle from 14\ser{\frac14} on fits under the curve from x=1x = 1.
14+19+116+⋯\displaystyle \ser{\frac14 + \frac19 + \frac1{16} + \cdots} ≤∫1∞dxx2\displaystyle \le \crv{\int_1^\infty \frac{dx}{x^2}} =1\displaystyle = \crv{1}
Add the first term backThe rectangle for 1\aon{1} sits over [0,1][0, 1], where the area under the curve is infinite, so we keep it separate.
1+14+19+116+⋯\displaystyle \ser{1 + \frac14 + \frac19 + \frac1{16} + \cdots} ≤1+1\displaystyle \le \aon{1} + \crv{1} =2\displaystyle = 2
Bounded, so it convergesThe running total only increases and never passes 22, so the series converges. Its sum is about 1.645\ser{1.645}, which is not the integral's value 1\crv{1}.
∑n=1∞1n2≈1.645\displaystyle \ser{\sum_{n=1}^{\infty}\frac{1}{n^2}} \approx \ser{1.645}
∫1∞dxx2\displaystyle \crv{\int_1^\infty \frac{dx}{x^2}} =1\displaystyle = \crv{1}
2

Why the harmonic series diverges

Rectangles under y=1x\crv{y = \frac1x} don't help with the harmonic series. The area under that curve is infinite, so they only show the sum is at most infinity. To prove it diverges, we need a lower bound that goes to infinity, so we slide every rectangle 11 unit to the right, until it sits over the curve.

00
11
22
33
44
55
66
77
xx
yy
1\htmlClass{lc-blue}{1}
12\htmlClass{lc-blue}{\dfrac{1}{2}}
13\htmlClass{lc-blue}{\dfrac{1}{3}}
14\htmlClass{lc-blue}{\dfrac{1}{4}}
15\htmlClass{lc-blue}{\dfrac{1}{5}}
16\htmlClass{lc-blue}{\dfrac{1}{6}}
ln⁡7\htmlClass{lc-pink}{\ln 7}
y=1x\htmlClass{lc-pink}{y = \dfrac{1}{x}}
1x\displaystyle \crv{\frac{1}{x}} ≤1nfor n≤x≤n+1\displaystyle \le \ser{\frac{1}{n}} \quad {\text{for } n \le x \le n + 1}
1+12+⋯+1N≥∫1N+1dxx=ln⁡(N+1)\displaystyle \begin{aligned}\ser{1 + \frac12 + \cdots + \frac1N} &\ge \crv{\int_1^{N+1} \frac{dx}{x}}\\ &= \crv{\ln(N+1)}\end{aligned}
Show that 1+12+13+⋯\ser{1 + \frac12 + \frac13 + \cdots} diverges by comparing it with ∫1∞dxx\crv{\int_1^\infty \frac{dx}{x}}
00
11
22
33
44
55
66
77
88
yy
y=1x\htmlClass{lc-pink}{y = \dfrac{1}{x}}
Text solution
An upper bound says nothingWith the rectangles under the curve, the bound is infinity, and every sum of positive numbers is at most infinity.
1+12+13+⋯\displaystyle \ser{1 + \frac12 + \frac13 + \cdots} ≤1+∫1∞dxx\displaystyle \le \aon{1} + \crv{\int_1^\infty \frac{dx}{x}} =∞\displaystyle = \crv{\infty}
Slide the rectangles over the curveOn [n,n+1][n, n + 1], the decreasing curve is highest at the left edge, where it equals 1n\ser{\frac1n}. So the first NN rectangles cover the area from x=1x = 1 to x=N+1x = N + 1.
1+12+⋯+1N\displaystyle \ser{1 + \frac12 + \cdots + \frac1N} ≥∫1N+1dxx\displaystyle \ge \crv{\int_1^{N+1} \frac{dx}{x}} =ln⁡(N+1)\displaystyle = \crv{\ln(N+1)}
The lower bound goes to infinityAs NN goes to infinity, ln⁡(N+1)\crv{\ln(N + 1)} does too, so the series diverges. It grows like the natural log, which is why the running totals climb so slowly.
1+12+⋯+1100≈5.19\displaystyle \ser{1 + \frac12 + \cdots + \frac{1}{100}} \approx \ser{5.19} ≥ln⁡101≈4.62\displaystyle \ge \crv{\ln 101} \approx \crv{4.62}

So rectangles under the curve give an upper bound, and rectangles over it give a lower bound. Which comparison proves convergence, and which proves divergence?

For any positive, decreasing function f\crv{f} with an=f(n)\ser{a_n = f(n)}, the two placements put the sum between ∫1∞f(x) dx\crv{\int_1^\infty f(x)\,dx} and a1+∫1∞f(x) dx\aon{a_1} + \crv{\int_1^\infty f(x)\,dx}. If the integral is finite, the upper bound keeps the sum finite. If it's infinite, the lower bound forces the sum to infinity. That's the integral test.

f positive, continuous, decreasing for x≥1,an=f(n)\displaystyle {\crv{f} \text{ positive, continuous, decreasing for } x \ge 1, \quad \ser{a_n = f(n)}}
f positive, continuous,decreasing for x≥1,an=f(n)\displaystyle \begin{gathered}\crv{f} \text{ positive, continuous,}\\ \text{decreasing for } x \ge 1,\\ \ser{a_n = f(n)}\end{gathered}
∑n=1∞an and ∫1∞f(x) dx both converge or both diverge\displaystyle {\ser{\sum_{n=1}^{\infty} a_n} \text{ and } \crv{\int_1^\infty f(x)\,dx} \text{ both converge or both diverge}}
∑n=1∞an and ∫1∞f(x) dxboth converge or both diverge\displaystyle \begin{gathered}\ser{\sum_{n=1}^{\infty} a_n} \text{ and } \crv{\int_1^\infty f(x)\,dx}\\ \text{both converge or both diverge}\end{gathered}
∫1∞f(x) dx\displaystyle \crv{\int_1^\infty f(x)\,dx} ≤∑n=1∞an\displaystyle \le \ser{\sum_{n=1}^{\infty} a_n} ≤a1+∫1∞f(x) dx\displaystyle \le \aon{a_1} + \crv{\int_1^\infty f(x)\,dx}

The test needs f\crv{f} to be decreasing so that f(n)\ser{f(n)} is the lowest value on [n−1,n][n - 1, n] and the highest value on [n,n+1][n, n + 1]. It needs f\crv{f} to be positive so the running totals only increase, and continuous so the integral is defined.

Without decreasing, the heights at whole numbers tell us nothing about the area between them. For f(x)=1x2+sin⁡2(πx)\crv{f(x) = \frac{1}{x^2} + \sin^2(\pi x)}, every f(n)\ser{f(n)} is 1n2\ser{\frac{1}{n^2}}, so the series converges, but each bump adds an area of 12\crv{\frac12}, so the integral is infinite.

00
11
22
33
44
55
66
xx
yy
12\htmlClass{lc-pink}{\dfrac{1}{2}}
y=1x2+sin⁡2(πx)\htmlClass{lc-pink}{y = \dfrac{1}{x^2} + \sin^2(\pi x)}
y=1x2\htmlClass{lc-mute}{y = \dfrac{1}{x^2}}
f(n)=1n2\displaystyle \ser{f(n) = \frac{1}{n^2}}
∑n=1∞f(n)≈1.645\displaystyle {\ser{\sum_{n=1}^{\infty} f(n)} \approx \ser{1.645}}
∫nn+1sin⁡2(πx) dx=12\displaystyle \crv{\int_n^{n+1} \sin^2(\pi x)\,dx = \frac12}
∫1∞f(x) dx=∞\displaystyle \crv{\int_1^\infty f(x)\,dx = \infty}
3

Checking that the function is decreasing

Let's use the test on e−1+2e−4+3e−9+⋯\ser{e^{-1} + 2e^{-4} + 3e^{-9} + \cdots}. Its function is f(x)=xe−x2\crv{f(x) = x e^{-x^2}}, and here it isn't obvious that f\crv{f} is decreasing: the factor xx grows while e−x2e^{-x^2} shrinks. So we check the sign of f′(x)f'(x) before integrating.

Does ∑n=1∞ne−n2\ser{\sum_{n=1}^{\infty} n e^{-n^2}} converge or diverge?
∑n=1∞ne−n2\displaystyle \ser{\sum_{n=1}^{\infty} n e^{-n^2}} =e−1+2e−4+3e−9+⋯\displaystyle = \ser{e^{-1} + 2e^{-4} + 3e^{-9} + \cdots}
converges or diverges?\displaystyle \text{converges or diverges?}
Text solution
Positive and decreasingFor x≥1x \ge 1, both factors are positive. And 2x2≥22x^2 \ge 2, so 1−2x21 - 2x^2 is negative and f\crv{f} is decreasing.
f′(x)=e−x2+x⋅(−2x)e−x2=e−x2(1−2x2)<0for x≥1\displaystyle {f'(x) = e^{-x^2} + x\cdot(-2x)e^{-x^2} = e^{-x^2}(1 - 2x^2) < 0 \quad \text{for } x \ge 1}
f′(x)=e−x2+x⋅(−2x)e−x2=e−x2(1−2x2)<0for x≥1\displaystyle \begin{aligned}f'(x) &= e^{-x^2} + x\cdot(-2x)e^{-x^2}\\ &= e^{-x^2}(1 - 2x^2)\\ &< 0 \quad \text{for } x \ge 1\end{aligned}
Integrate with u=x2u = x^2Then du=2x dxdu = 2x\,dx, so x dxx\,dx is 12 du\frac12\,du.
∫1∞xe−x2 dx\displaystyle \crv{\int_1^\infty x e^{-x^2}\,dx} =lim⁡b→∞[−12e−x2]1b\displaystyle = \lim_{b\to\infty}\Big[-\frac12 e^{-x^2}\Big]_1^b =12e\displaystyle = \crv{\frac{1}{2e}}
∫1∞xe−x2 dx=lim⁡b→∞[−12e−x2]1b=12e\displaystyle \begin{aligned}\crv{\int_1^\infty x e^{-x^2}\,dx} &= \lim_{b\to\infty}\Big[-\frac12 e^{-x^2}\Big]_1^b\\ &= \crv{\frac{1}{2e}}\end{aligned}
The series convergesThe integral is finite, so the series converges. Its sum is about 0.405\ser{0.405}, not 12e≈0.184\crv{\frac{1}{2e} \approx 0.184}. The test tells us whether a series converges, not what it adds up to.
∑n=1∞ne−n2converges\displaystyle \ser{\sum_{n=1}^{\infty} n e^{-n^2}} \quad \cvg{\text{converges}}
4

When the terms increase before they decrease

Now let's look at the sum of ln⁡nn\ser{\frac{\ln n}{n}}. Its first few terms go up before they come down:

ln⁡11=0\displaystyle \ser{\frac{\ln 1}{1} = 0}
ln⁡22≈0.347\displaystyle \ser{\frac{\ln 2}{2} \approx 0.347}
ln⁡33≈0.366\displaystyle \ser{\frac{\ln 3}{3} \approx 0.366}
ln⁡44≈0.347\displaystyle \ser{\frac{\ln 4}{4} \approx 0.347}
ln⁡55≈0.322\displaystyle \ser{\frac{\ln 5}{5} \approx 0.322}

The integral test needs a decreasing function, so we find where f(x)=ln⁡xx\crv{f(x) = \frac{\ln x}{x}} starts decreasing and set aside the terms before that.

Determine whether ∑n=1∞ln⁡nn\ser{\sum_{n=1}^{\infty}\frac{\ln n}{n}} converges or diverges
∑n=1∞ln⁡nn\displaystyle \ser{\sum_{n=1}^{\infty}\frac{\ln n}{n}}
converges or diverges?\displaystyle \text{converges or diverges?}
Text solution
Find where f\crv{f} decreasesThe denominator x2x^2 is positive, so f′(x)<0f'(x) < 0 when ln⁡x>1\ln x > 1, that is, for x>ex > e. Since ee is between 22 and 33, f\crv{f} decreases for all x≥3x \ge 3.
f′(x)=x⋅1x−ln⁡xx2=1−ln⁡xx2<0for x>e≈2.72\displaystyle {f'(x) = \frac{x\cdot\frac1x - \ln x}{x^2} = \frac{1 - \ln x}{x^2} < 0 \quad \text{for } x > e \approx 2.72}
f′(x)=x⋅1x−ln⁡xx2=1−ln⁡xx2<0for x>e≈2.72\displaystyle \begin{aligned}f'(x) &= \frac{x\cdot\frac1x - \ln x}{x^2}\\ &= \frac{1 - \ln x}{x^2}\\ &< 0 \quad \text{for } x > e \approx 2.72\end{aligned}
Split off the first two termsThey add up to about 0.3470.347, a finite number, so they don't change whether the series converges.
∑n=1∞ln⁡nn\displaystyle \ser{\sum_{n=1}^{\infty}\frac{\ln n}{n}} =0+ln⁡22+∑n=3∞ln⁡nn\displaystyle = \mut{0 + \frac{\ln 2}{2}} + \ser{\sum_{n=3}^{\infty}\frac{\ln n}{n}}
Use the test from 33With u=ln⁡xu = \ln x and du=dxxdu = \frac{dx}{x}, the integral becomes ∫u du\int u\,du. It goes to infinity, so the series from n=3n = 3 diverges, and adding the first two terms back keeps it infinite.
∫3∞ln⁡xx dx\displaystyle \crv{\int_3^\infty \frac{\ln x}{x}\,dx} =lim⁡b→∞(12(ln⁡b)2−12(ln⁡3)2)\displaystyle = \lim_{b\to\infty}\crv{\Big(\frac12(\ln b)^2 - \frac12(\ln 3)^2\Big)} =∞\displaystyle = \crv{\infty}
∫3∞ln⁡xx dx=lim⁡b→∞(12(ln⁡b)2−12(ln⁡3)2)=∞\displaystyle \begin{aligned}&\crv{\int_3^\infty \frac{\ln x}{x}\,dx}\\ &\quad= \lim_{b\to\infty}\crv{\Big(\frac12(\ln b)^2 - \frac12(\ln 3)^2\Big)}\\ &\quad= \crv{\infty}\end{aligned}
∑n=1∞ln⁡nndiverges\displaystyle \ser{\sum_{n=1}^{\infty}\frac{\ln n}{n}} \quad \dvg{\text{diverges}}
5

The p-series

We've seen that ∑1n2\ser{\sum \frac{1}{n^2}} converges, while ∑1n\ser{\sum \frac1n} diverges. In one the power on nn is 22, and in the other it's 11. So at what power pp does the series switch from diverging to converging?

For which values of pp does ∑n=1∞1np\ser{\sum_{n=1}^{\infty}\frac{1}{n^p}} converge?
∑n=1∞1np\displaystyle \ser{\sum_{n=1}^{\infty}\frac{1}{n^p}} =1+12p+13p+⋯\displaystyle = \ser{1 + \frac{1}{2^p} + \frac{1}{3^p} + \cdots}
converges for p\displaystyle \text{converges for } p =  ?\displaystyle = \;?
diverges for p\displaystyle \text{diverges for } p =  ?\displaystyle = \;?
Text solution
Use the integral test for p>0p > 0Then 1xp\crv{\frac{1}{x^p}} is positive and decreasing, so the series goes with ∫1∞dxxp\crv{\int_1^\infty \frac{dx}{x^p}}.
∫1bdxxp=b1−p−11−p(p≠1)∫1bdxx=ln⁡b\displaystyle \begin{aligned}\crv{\int_1^b \frac{dx}{x^p}} &= \crv{\frac{b^{1-p} - 1}{1 - p}} \quad (p \ne 1)\\[12pt] \crv{\int_1^b \frac{dx}{x}} &= \crv{\ln b}\end{aligned}
Let bb go to infinityWhen p>1p > 1, the exponent 1−p1 - p is negative and the integral is finite. When p<1p < 1, b1−pb^{1-p} goes to infinity, and when p=1p = 1, ln⁡b\crv{\ln b} does too.
p>1 ⁣: b1−p→0\displaystyle \cvg{p > 1\colon\ b^{1-p} \to 0}
p<1 ⁣: b1−p→∞\displaystyle \dvg{p < 1\colon\ b^{1-p} \to \infty}
When p≤0p \le 0Now 1xp\crv{\frac{1}{x^p}} isn't decreasing, but the terms don't go to 00 (for p=0p = 0 every term is 1\ser{1}), so the series diverges by the nth-term test.

We call ∑1np\ser{\sum \frac{1}{n^p}} a p-series. When p=1p = 1, it's the harmonic series.

∑n=1∞1np    {converges if p>1diverges if p≤1\displaystyle \ser{\sum_{n=1}^{\infty}\frac{1}{n^p}}\;\;\begin{cases}\cvg{\text{converges if } p > 1}\\[2pt] \dvg{\text{diverges if } p \le 1}\end{cases}

To test a series with roots or powers of nn, rewrite the term with a single power of nn in the denominator, then compare pp with 11.

1n=1n12\displaystyle \ser{\frac{1}{\sqrt n} = \frac{1}{n^{\frac12}}}
p=12≤1 ⁣: diverges\displaystyle \dvg{p = \frac12 \le 1\colon\ \text{diverges}}
nn3=n12n3=1n52\displaystyle \ser{\frac{\sqrt n}{n^3} = \frac{n^{\frac12}}{n^3} = \frac{1}{n^{\frac52}}}
p=52>1 ⁣: converges\displaystyle \cvg{p = \frac52 > 1\colon\ \text{converges}}
1n23=1n23\displaystyle \ser{\frac{1}{\sqrt[3]{n^2}} = \frac{1}{n^{\frac23}}}
p=23≤1 ⁣: diverges\displaystyle \dvg{p = \frac23 \le 1\colon\ \text{diverges}}
∑n=1∞5n3=5∑n=1∞1n3\displaystyle \ser{\sum_{n=1}^{\infty}\frac{5}{n^3} = 5\sum_{n=1}^{\infty}\frac{1}{n^3}}
p=3>1 ⁣: converges\displaystyle \cvg{p = 3 > 1\colon\ \text{converges}}
1n1.1\displaystyle \ser{\frac{1}{n^{1.1}}}
p=1.1>1 ⁣: converges\displaystyle \cvg{p = 1.1 > 1\colon\ \text{converges}}

Be careful about where nn is. When nn is in the exponent, as in 12n\ser{\frac{1}{2^n}}, the series is geometric, not a p-series.

12n=(12)n\displaystyle \ser{\frac{1}{2^n} = \Big(\frac12\Big)^n}
geometric, r\displaystyle \text{geometric}, \ r =12\displaystyle = \frac12
1n2\displaystyle \ser{\frac{1}{n^2}}
p-series, p\displaystyle \text{p-series}, \ p =2\displaystyle = 2
6

Estimating the sum of a series

If we add up the first 1010 terms of ∑n=1∞1n2\ser{\sum_{n=1}^\infty \frac{1}{n^2}}, we get about 1.54981.5498. How close is that to the whole sum? The error is everything we left out, the terms from 1112\ser{\frac{1}{11^2}} on.

∑n=1∞1n2−(1+14+⋯+1102)\displaystyle {\ser{\sum_{n=1}^{\infty}\frac{1}{n^2}} - \Big(\ser{1 + \frac14 + \cdots + \frac{1}{10^2}}\Big)} =1112+1122+1132+⋯=  ?\displaystyle = {\ser{\frac{1}{11^2} + \frac{1}{12^2} + \frac{1}{13^2} + \cdots} = \;?}

We can't add those by hand, but we can trap them between two areas, with the same rectangles we used for the integral test.

1010
1111
1212
1313
1414
1515
1616
1717
0.01\htmlClass{lc-mute}{0.01}
1112\htmlClass{lc-blue}{\dfrac{1}{11^2}}
1122\htmlClass{lc-blue}{\dfrac{1}{12^2}}
1132\htmlClass{lc-blue}{\dfrac{1}{13^2}}
1142\htmlClass{lc-blue}{\dfrac{1}{14^2}}
1152\htmlClass{lc-blue}{\dfrac{1}{15^2}}
1162\htmlClass{lc-blue}{\dfrac{1}{16^2}}
110\htmlClass{lc-pink}{\dfrac{1}{10}}
1010
1111
1212
1313
1414
1515
1616
1717
0.01\htmlClass{lc-mute}{0.01}
1112\htmlClass{lc-blue}{\dfrac{1}{11^2}}
1122\htmlClass{lc-blue}{\dfrac{1}{12^2}}
1132\htmlClass{lc-blue}{\dfrac{1}{13^2}}
1142\htmlClass{lc-blue}{\dfrac{1}{14^2}}
1152\htmlClass{lc-blue}{\dfrac{1}{15^2}}
111\htmlClass{lc-pink}{\dfrac{1}{11}}
∫10∞dxx2=110∫11∞dxx2=111\displaystyle \begin{aligned}\crv{\int_{10}^\infty \frac{dx}{x^2}} &= \crv{\frac{1}{10}}\\[6pt] \crv{\int_{11}^\infty \frac{dx}{x^2}} &= \crv{\frac{1}{11}}\end{aligned}
111\displaystyle \crv{\frac{1}{11}} ≤1112+1122+⋯\displaystyle \le \ser{\frac{1}{11^2} + \frac{1}{12^2} + \cdots} ≤110\displaystyle \le \crv{\frac{1}{10}}
Estimate ∑n=1∞1n2\ser{\sum_{n=1}^\infty \frac{1}{n^2}} using the first 10 terms
00
55
1010
1515
y=1x2\htmlClass{lc-pink}{y = \dfrac{1}{x^2}}
Text solution
Under the curve from 1010, over it from 1111Because 1x2\crv{\frac{1}{x^2}} is decreasing, the rectangle for 1112\ser{\frac{1}{11^2}} over [10,11][10, 11], and every one after it, lies under the curve. Slid 11 unit to the right, the same rectangles lie above it.
∫11∞dxx2\displaystyle \crv{\int_{11}^\infty \frac{dx}{x^2}} ≤1112+1122+⋯\displaystyle \le \ser{\frac{1}{11^2} + \frac{1}{12^2} + \cdots} ≤∫10∞dxx2\displaystyle \le \crv{\int_{10}^\infty \frac{dx}{x^2}}
Add back the first 1010 termsAdding 1.54981.5498 to both bounds traps the whole sum in an interval less than 0.010.01 wide. The true sum, about 1.64491.6449, is inside it.
1.5498+111≈1.6407\displaystyle 1.5498 + \crv{\frac{1}{11}} \approx 1.6407 ≤∑n=1∞1n2\displaystyle \le \ser{\sum_{n=1}^{\infty}\frac{1}{n^2}} ≤1.5498+110≈1.6498\displaystyle \le 1.5498 + \crv{\frac{1}{10}} \approx 1.6498
∑n=1∞1n2≥1.5498+111≈1.6407∑n=1∞1n2≤1.5498+110≈1.6498\displaystyle \begin{aligned}\ser{\sum_{n=1}^{\infty}\frac{1}{n^2}} &\ge 1.5498 + \crv{\frac{1}{11}} \approx 1.6407\\[6pt] \ser{\sum_{n=1}^{\infty}\frac{1}{n^2}} &\le 1.5498 + \crv{\frac{1}{10}} \approx 1.6498\end{aligned}

The same comparison works after any number of terms NN, for any positive, decreasing function f\crv{f}. The first missing term is f(N+1)\ser{f(N + 1)}: under the curve its rectangle is on [N,N+1][N, N + 1], and over the curve it's on [N+1,N+2][N + 1, N + 2].

∫N+1∞f(x) dx\displaystyle \crv{\int_{N+1}^\infty f(x)\,dx} ≤f(N+1)+f(N+2)+⋯\displaystyle \le \ser{f(N+1) + f(N+2) + \cdots} ≤∫N∞f(x) dx\displaystyle \le \crv{\int_N^\infty f(x)\,dx}

The upper bound tells us how large our error can be. For ∑n=1∞1n2\ser{\sum_{n=1}^\infty \frac{1}{n^2}}, the remaining terms add to at most 1N\crv{\frac1N}, so to keep the error at most 0.0010.001, we need 10001000 terms.

1(N+1)2+1(N+2)2+⋯\displaystyle \ser{\frac{1}{(N+1)^2} + \frac{1}{(N+2)^2} + \cdots} ≤∫N∞dxx2\displaystyle \le \crv{\int_N^\infty \frac{dx}{x^2}} =1N\displaystyle = \crv{\frac1N}
1N≤0.001  →  N=1000\displaystyle {\crv{\frac1N} \le 0.001 \;\to\; N = 1000}

Summary

The integral bounds the sum
00
11
22
33
44
00
11
22
33
44
The bounds
∫1∞f(x) dx\displaystyle \crv{\int_1^\infty f(x)\,dx} ≤∑n=1∞an\displaystyle \le \ser{\sum_{n=1}^{\infty} a_n} ≤a1+∫1∞f(x) dx\displaystyle \le \aon{a_1} + \crv{\int_1^\infty f(x)\,dx}
For 1n2\ser{\frac{1}{n^2}}
1\displaystyle \crv{1} ≤∑n=1∞1n2\displaystyle \le \ser{\sum_{n=1}^{\infty}\frac{1}{n^2}} ≤1+1\displaystyle \le \aon{1} + \crv{1} =2\displaystyle = 2
∑n=1∞1n2≈1.645\displaystyle \ser{\sum_{n=1}^{\infty}\frac{1}{n^2} \approx 1.645}

When f\crv{f} is positive and decreasing, we use rectangles of width 11 under the curve to show that the sum is at most a1+∫1∞f(x) dx\aon{a_1} + \crv{\int_1^\infty f(x)\,dx}, and rectangles over the curve to show that it is at least ∫1∞f(x) dx\crv{\int_1^\infty f(x)\,dx}. Because the series is between these two numbers, its sum is finite exactly when the improper integral is finite. The integral gives bounds on the sum, not the sum itself. For ∑1n2\ser{\sum \frac{1}{n^2}}, the integral is 1\crv{1}, but the sum is about 1.645\ser{1.645}.

The integral test
The test
f positive, continuous, decreasing for x≥1,an=f(n)∑n=1∞an and ∫1∞f(x) dx both converge or both diverge\displaystyle \begin{gathered}\crv{f} \text{ positive, continuous, decreasing for } x \ge 1, \quad \ser{a_n = f(n)}\\ \ser{\sum_{n=1}^{\infty} a_n} \text{ and } \crv{\int_1^\infty f(x)\,dx} \text{ both converge or both diverge}\end{gathered}
f positive, continuous,decreasing for x≥1,an=f(n)∑n=1∞an and ∫1∞f(x) dxboth converge or both diverge\displaystyle \begin{gathered}\crv{f} \text{ positive, continuous,}\\ \text{decreasing for } x \ge 1,\\ \ser{a_n = f(n)}\\[6pt] \ser{\sum_{n=1}^{\infty} a_n} \text{ and } \crv{\int_1^\infty f(x)\,dx}\\ \text{both converge or both diverge}\end{gathered}
Decreasing from n=3n = 3 on
ln⁡nn ⁣:  f′(x)=1−ln⁡xx2<0 for x>e→  ∑n=3∞ln⁡nn, ∫3∞ln⁡xx dx\displaystyle {\ser{\frac{\ln n}{n}}\colon\; f'(x) = \frac{1 - \ln x}{x^2} < 0 \text{ for } x > e \quad\to\; \ser{\sum_{n=3}^{\infty}\frac{\ln n}{n}}, \ \crv{\int_3^\infty \frac{\ln x}{x}\,dx}}
ln⁡nn ⁣:  f′(x)=1−ln⁡xx2<0for x>e→  ∑n=3∞ln⁡nn, ∫3∞ln⁡xx dx\displaystyle \begin{gathered}\ser{\frac{\ln n}{n}}\colon\; f'(x) = \frac{1 - \ln x}{x^2} < 0\\ \text{for } x > e\\ \to\; \ser{\sum_{n=3}^{\infty}\frac{\ln n}{n}}, \ \crv{\int_3^\infty \frac{\ln x}{x}\,dx}\end{gathered}

Before running the integral test, check whether the terms approach 00. If lim⁡n→∞an≠0\lim_{n\to\infty} \ser{a_n} \ne 0, like n2n+1→12\ser{\frac{n}{2n+1}} \to \ser{\frac12}, the series diverges by the nth-term test, and no integral is needed. To apply the test, f\crv{f} must be positive, continuous, and decreasing. If it isn't clear whether f\crv{f} is decreasing, check the sign of f′(x)f'(x). For ln⁡nn\ser{\frac{\ln n}{n}}, f′(x)<0f'(x) < 0 only once x>ex > e, so we set aside the first two terms and start the integral test from n=3n = 3. The first few terms add only a finite amount, so setting them aside doesn't change whether the series converges.

Testing a p-series
The p-series rule
∑n=1∞1np    {converges if p>1diverges if p≤1\displaystyle \ser{\sum_{n=1}^{\infty}\frac{1}{n^p}}\;\;\begin{cases}\cvg{\text{converges if } p > 1}\\[2pt] \dvg{\text{diverges if } p \le 1}\end{cases}
Rewriting terms
1nn=1n32\displaystyle \ser{\frac{1}{n\sqrt n} = \frac{1}{n^{\frac32}}}
12n=(12)n\displaystyle \ser{\frac{1}{2^n} = \Big(\frac12\Big)^n}

A p-series ∑1np\ser{\sum \frac{1}{n^p}} converges if p>1\cvg{p > 1} and diverges if p≤1\dvg{p \le 1}. When p=1p = 1, we get the harmonic series 1+12+13+⋯\ser{1 + \frac12 + \frac13 + \cdots}, which diverges to infinity. Before checking pp, rewrite the denominator as a single power of nn, like turning 1nn\ser{\frac{1}{n\sqrt n}} into 1n32\ser{\frac{1}{n^{\frac32}}}. If nn appears in the exponent instead of the base, like 12n=(12)n\ser{\frac{1}{2^n} = \left(\frac12\right)^n}, it's a geometric series, not a p-series.

Bounding the error after NN terms
Bounds on the remaining terms
∫N+1∞f(x) dx\displaystyle \crv{\int_{N+1}^\infty f(x)\,dx} ≤f(N+1)+f(N+2)+⋯\displaystyle \le \ser{f(N+1) + f(N+2) + \cdots} ≤∫N∞f(x) dx\displaystyle \le \crv{\int_N^\infty f(x)\,dx}

After adding the first NN terms of a series with f\crv{f} positive and decreasing, the sum of the remaining terms is between ∫N+1∞f(x) dx\crv{\int_{N+1}^\infty f(x)\,dx} and ∫N∞f(x) dx\crv{\int_N^\infty f(x)\,dx}. The upper bound tells us how large the error can be. For ∑n=1∞1n3\ser{\sum_{n=1}^\infty \frac{1}{n^3}} after 55 terms, the error is at most ∫5∞dxx3=150\crv{\int_5^\infty \frac{dx}{x^3} = \frac{1}{50}}.

7

Practice

1
Which of these series converge? I. ∑n=1∞1ne\displaystyle\sum_{n=1}^{\infty}\frac{1}{n^e} II. ∑n=1∞nn\displaystyle\sum_{n=1}^{\infty}\frac{\sqrt n}{n} III. ∑n=1∞en\displaystyle\sum_{n=1}^{\infty}\frac{e}{n}

Write each term in the form cnp\frac{c}{n^p}. What is pp for each series?

If you got (C) In series III, ee is a constant factor in the numerator, not the exponent on nn.

If you got (B) For series II, simplify nn\frac{\sqrt n}{n} into 1np\frac{1}{n^p}. What is pp?

In series I, ee is in the exponent, so p=e\cvg{p = e}. Since e≈2.72>1\cvg{e \approx 2.72 > 1}, this series converges.
I: p=e≈2.72>1\displaystyle \text{I: } \cvg{p = e \approx 2.72 > 1}
In series II, dividing n12\ser{n^{\frac12}} by n1\ser{n^1} leaves n12\ser{n^{\frac12}} in the denominator, so p=12≤1\dvg{p = \frac12 \le 1} and the series diverges.
II: nn=n12n=1n12,p=12≤1\displaystyle \text{II: } \ser{\frac{\sqrt n}{n} = \frac{n^{\frac12}}{n} = \frac{1}{n^{\frac12}}}, \qquad \dvg{p = \frac12 \le 1}
In series III, ee is a constant factor, not an exponent. Pulling ee out leaves the harmonic series, which diverges.
III: ∑n=1∞en=e∑n=1∞1n,p=1\displaystyle \text{III: } \ser{\sum_{n=1}^{\infty}\frac{e}{n} = e\sum_{n=1}^{\infty}\frac1n}, \qquad \dvg{p = 1}
2
Use the integral test to determine whether ∑n=1∞ln⁡nn2\displaystyle\sum_{n=1}^{\infty}\frac{\ln n}{n^2} converges or diverges.

Find the derivative of f(x)=ln⁡xx2f(x) = \frac{\ln x}{x^2}. For which values of xx is f′(x)<0f'(x) < 0?

Find where f\crv{f} is decreasingUse the quotient rule to find where f′(x)f'(x) is negative. Here f′(x)<0f'(x) < 0 for x>e≈1.65x > \sqrt{e} \approx 1.65, so f\crv{f} is decreasing for x≥2x \ge 2. That lets us use the integral test starting at n=2n = 2, and leaving out f(1)=0\ser{f(1) = 0} doesn't change whether the series converges.
f′(x)\displaystyle f'(x) =x2⋅1x−2xln⁡xx4\displaystyle = \frac{x^2\cdot\frac1x - 2x\ln x}{x^4} =1−2ln⁡xx3<0for x>e≈1.65\displaystyle = \frac{1 - 2\ln x}{x^3} < 0 \quad {\text{for } x > \sqrt e \approx 1.65}
Integrate by partsLet u=ln⁡xu = \ln x and dv=dxx2dv = \frac{dx}{x^2} to find the antiderivative.
u=ln⁡x,  dv=dxx2,du=dxx,  v=−1x\displaystyle {u = \ln x,\ \ dv = \frac{dx}{x^2}}, \qquad {du = \frac{dx}{x},\ \ v = -\frac1x}
∫ln⁡xx2 dx\displaystyle \int \frac{\ln x}{x^2}\,dx =−ln⁡xx+∫dxx2\displaystyle = -\frac{\ln x}{x} + \int \frac{dx}{x^2} =−ln⁡xx−1x+C\displaystyle = -\frac{\ln x}{x} - \frac1x + C
Evaluate from 22 to infinityBy L'Hôpital's rule, ln⁡bb\frac{\ln b}{b} goes to 00 as b→∞b \to \infty. The integral is finite, so the series converges.
∫2∞ln⁡xx2 dx\displaystyle \crv{\int_2^\infty \frac{\ln x}{x^2}\,dx} =lim⁡b→∞(−ln⁡bb−1b)+ln⁡22+12\displaystyle = \lim_{b\to\infty}\Big(-\frac{\ln b}{b} - \frac1b\Big) + \frac{\ln 2}{2} + \frac12 =ln⁡22+12≈0.85\displaystyle = \crv{\frac{\ln 2}{2} + \frac12} \approx \crv{0.85}
∑n=1∞ln⁡nn2converges\displaystyle \ser{\sum_{n=1}^{\infty}\frac{\ln n}{n^2}} \quad \cvg{\text{converges}}
3
The function ff is positive, continuous and decreasing for x≥1x \ge 1, with an=f(n)a_n = f(n) and ∫1∞f(x) dx=3\displaystyle\int_1^\infty f(x)\,dx = 3. Which of the following must be true?

If each rectangle has width 11 and height an=f(n)a_n = f(n) over the interval [n,n+1][n, n + 1], does it lie above or below the curve?

If you got (A) The rectangles extend above the curve, so their total area cannot equal the area under the curve.

If you got (C) Check which rectangles lie under the curve. Rectangles beneath the curve only tell us a2+a3+⋯≤3a_2 + a_3 + \dots \le 3. Since a1a_1 is missing from that inequality, the full sum doesn't have to be less than 33.

Since f\crv{f} is decreasing, each rectangle of height an=f(n)\ser{a_n = f(n)} over [n,n+1][n, n+1] lies completely above the curve, so the sum of their areas is strictly greater than the integral.
n<x≤n+1:an=f(n)>f(x)  →  ∑n=1∞an>∫1∞f(x) dx=3\displaystyle {n < x \le n + 1: \quad \ser{a_n = f(n)} > \crv{f(x)} \;\to\; \ser{\sum_{n=1}^{\infty} a_n} > \crv{\int_1^\infty f(x)\,dx = 3}}
n<x≤n+1:an=f(n)>f(x)→  ∑n=1∞an>∫1∞f(x) dx=3\displaystyle \begin{gathered}n < x \le n + 1: \quad \ser{a_n = f(n)} > \crv{f(x)}\\ \to\; \ser{\sum_{n=1}^{\infty} a_n} > \crv{\int_1^\infty f(x)\,dx = 3}\end{gathered}
To see that the sum is finite, set aside the first term and put each remaining rectangle for an\ser{a_n} over [n−1,n][n - 1, n] under the curve. That gives a2+a3+…≤3\ser{a_2 + a_3 + \dots} \le \crv{3}, so adding a1\aon{a_1} gives ∑n=1∞an≤a1+3\ser{\sum_{n=1}^\infty a_n} \le \aon{a_1} + \crv{3}.
∑n=1∞an\displaystyle \ser{\sum_{n=1}^{\infty} a_n} ≤a1+∫1∞f(x) dx\displaystyle \le \aon{a_1} + \crv{\int_1^\infty f(x)\,dx} =a1+3\displaystyle = \aon{a_1} + \crv{3}
4
The sum of the first 5 terms of ∑n=1∞1n3\displaystyle\sum_{n=1}^{\infty}\frac{1}{n^3} is used to approximate the sum of the series. Find an upper bound for the error.

To bound the remaining terms from n=6n = 6 onward, place each rectangle under the curve. Which interval does the first rectangle lie over?

The error is the sum of all the terms left out after the first 55 terms.
error\displaystyle \ser{\text{error}} =163+173+183+⋯\displaystyle = \ser{\frac{1}{6^3} + \frac{1}{7^3} + \frac{1}{8^3} + \cdots}
The rectangle for 163\ser{\frac{1}{6^3}} lies under the curve on [5,6][5, 6], so the sum of the remaining terms is at most the integral from 55 to infinity.
163+173+⋯\displaystyle \ser{\frac{1}{6^3} + \frac{1}{7^3} + \cdots} ≤∫5∞dxx3\displaystyle \le \crv{\int_5^\infty \frac{dx}{x^3}} =lim⁡b→∞[−12x2]5b\displaystyle = \lim_{b\to\infty}\Big[-\frac{1}{2x^2}\Big]_5^b =12⋅25\displaystyle = \frac{1}{2\cdot 25} =150\displaystyle = \crv{\frac{1}{50}}
5
For which values of pp does ∑n=2∞1n(ln⁡n)p\displaystyle\sum_{n=2}^{\infty}\frac{1}{n(\ln n)^p} converge? (Challenge)

Use the substitution u=ln⁡xu = \ln x on the integral ∫2∞dxx(ln⁡x)p\int_2^\infty \frac{dx}{x(\ln x)^p}. What integral in uu do you get?

If you got p≥1p \ge 1 Check the case p=1p = 1, which gives the series ∑1nln⁡n\sum \frac{1}{n\ln n}. Does that integral converge?

Substitute u=ln⁡xu = \ln xLet u=ln⁡xu = \ln x, so du=dxxdu = \frac{dx}{x}. When x=2x = 2, u=ln⁡2u = \ln 2, and as x→∞x \to \infty, u→∞u \to \infty.
u=ln⁡x,  du=dxx:∫2∞dxx(ln⁡x)p\displaystyle {u = \ln x,\ \ du = \frac{dx}{x}:} \qquad \crv{\int_2^\infty \frac{dx}{x(\ln x)^p}} =∫ln⁡2∞duup\displaystyle = \crv{\int_{\ln 2}^\infty \frac{du}{u^p}}
Apply the p-integral ruleThe integral of 1up\crv{\frac{1}{u^p}} from ln⁡2\ln 2 to infinity converges if p>1\cvg{p > 1} and diverges if p≤1\dvg{p \le 1}.
∫ln⁡2∞duup: finite if p>1, ∞ if p≤1\displaystyle \crv{\int_{\ln 2}^\infty \frac{du}{u^p}}: \ \cvg{\text{finite if } p > 1}, \ \dvg{\infty \text{ if } p \le 1}
Check negative values of ppWhen p<0p < 0, f(x)\crv{f(x)} is decreasing once ln⁡x>−p\ln x > -p, and the integral still diverges to infinity. So the series converges only for p>1\cvg{p > 1}.
p<0:f′(x)\displaystyle {p < 0:} \quad f'(x) =(ln⁡x)−p−1(−p−ln⁡x)x2<0for ln⁡x>−p\displaystyle = \frac{(\ln x)^{-p-1}(-p - \ln x)}{x^2} < 0 \quad {\text{for } \ln x > -p}