Topic 18

Convergence Tests

Start with the quickest check: if an↛0a_n \not\to 0, the series . If an→0a_n \to 0, let the terms pick the test. Powers of nn: compare with a pp-series ∑1np\sum \dfrac{1}{n^{p}}, which only for p>1p > 1. Factorials and cnc^{n}: the ratio test. A function you can integrate: the integral test. A sign that flips: the alternating series test. Always name the test you used; an answer with no reason earns few marks.

6 kinds of question. Open the one you are stuck on.

1

Show a series diverges using the divergence test

e.g. “Does the series converge or diverge? (the terms do not shrink to 00)”
  1. 1
    Find lim⁡n→∞an\lim_{n\to\infty} a_n

    Divide by the highest power, or use a limit you know.

  2. 2
    Compare the limit with 00

    Not 00, or no limit at all: the test applies. Exactly 00: the test says nothing.

  3. 3
    State the verdict and name the test

    A limit of 00 never proves convergence; move on to another test.

Remember: Terms must shrink to 0, or the series diverges.

See it in a problem4 worked examples

Does ∑n=1∞4n2−n7n2+2\displaystyle\sum_{n=1}^{\infty} \dfrac{4n^{2} - n}{7n^{2} + 2} converge or diverge?

  1. 1
    Find lim⁡n→∞an\lim_{n\to\infty} a_n

    lim⁡n→∞4−1/n7+2/n2=47\lim_{n\to\infty} \dfrac{4 - 1/n}{7 + 2/n^{2}} = \dfrac{4}{7}.

  2. 2
    Compare the limit with 00

    47≠0\dfrac{4}{7} \neq 0.

  3. 3
    State the verdict and name the test

    Diverges by the divergence test.

Answer:

Diverges, by the divergence test (an→47a_n \to \tfrac{4}{7}).

Common mistakes2
  • Saying "an→0a_n \to 0, so the series converges". The harmonic series ∑1n\sum \dfrac{1}{n} has an→0a_n \to 0 and diverges.
  • Taking the limit of the instead of the terms. This test is about ana_n.
2

Use the integral test to decide if a series converges

e.g. “Terms like 1n(log⁡n)p\dfrac{1}{n(\log n)^{p}} or nen2\dfrac{n}{e^{n^{2}}}, where f(x)f(x) has an easy antiderivative”
  1. 1
    Write f(x)f(x) with f(n)=anf(n) = a_n
  2. 2
    Check ff is positive, continuous and decreasing

    Decreasing for x≥Nx \geq N is enough. Show f′(x)<0f'(x) < 0. Or, if the top is a positive constant, show the bottom grows.

  3. 3
    Evaluate ∫N∞f(x) dx\int_N^{\infty} f(x)\,dx

    It is an improper integral: integrate to tt, then let t→∞t \to \infty.

  4. 4
    The series does what the integral does

Remember: The sum and the integral share one fate.

See it in a problem3 worked examples

Does ∑n=1∞nen2\displaystyle\sum_{n=1}^{\infty} \dfrac{n}{e^{n^{2}}} converge or diverge?

  1. 1
    Write f(x)f(x) with f(n)=anf(n) = a_n

    f(x)=xe−x2f(x) = x e^{-x^{2}}.

  2. 2
    Check ff is positive, continuous and decreasing

    f>0f > 0 and continuous for x≥1x \geq 1; f′(x)=e−x2(1−2x2)<0f'(x) = e^{-x^{2}}(1 - 2x^{2}) < 0 for x≥1x \geq 1.

  3. 3
    Evaluate ∫N∞f(x) dx\int_N^{\infty} f(x)\,dx

    ∫1∞xe−x2 dx=lim⁡t→∞[−12e−x2]1t=12e\int_1^{\infty} x e^{-x^{2}}\,dx = \lim_{t\to\infty}\left[-\tfrac{1}{2}e^{-x^{2}}\right]_1^{t} = \dfrac{1}{2e}.

  4. 4
    The series does what the integral does

    The integral converges, so the series converges.

Answer:

Converges, by the integral test.

Common mistakes3
  • Skipping the conditions. ff must be positive and decreasing; one that wiggles, like a function with sin⁡x\sin x in it, can break the test.
  • Saying the sum equals the integral. They converge together, but to different numbers.
  • Integrating a pp-series: ∑1np\sum \dfrac{1}{n^{p}} converges only for p>1p > 1; just quote that.
3

Use the comparison or limit comparison test

e.g. “A fraction of powers and roots of nn, like n+2n2+5\dfrac{\sqrt{n} + 2}{n^{2} + 5}”
  1. 1
    Keep only the dominant terms to build bnb_n

    Keep the biggest power on top and on the bottom; cnc^{n} with c>1c > 1 beats any power of nn. Drop a sin⁡n\sin n or cos⁡n\cos n only when a bigger term sits next to it. Next to a constant, as in 2+cos⁡nn3\dfrac{2 + \cos n}{n^{3}}, use ∣cos⁡n∣≤1|\cos n| \leq 1 and direct comparison. There anbn\dfrac{a_n}{b_n} has no limit.

  2. 2
    Decide ∑bn\sum b_n: a pp-series or a geometric series
  3. 3
    Compare: compute L=lim⁡anbnL = \lim \dfrac{a_n}{b_n}, or prove an inequality

    Limit comparison needs 0<L<∞0 < L < \infty. Direct comparison needs 0≤an≤bn0 \leq a_n \leq b_n with ∑bn\sum b_n convergent, or an≥bn≥0a_n \geq b_n \geq 0 with ∑bn\sum b_n divergent. Use ∣cos⁡n∣≤1|\cos n| \leq 1, log⁡n<n\log n < n.

  4. 4
    Conclude: ∑an\sum a_n does what ∑bn\sum b_n does

    Name the test you used.

Remember: Big terms decide; the comparison makes it official.

See it in a problem4 worked examples

Does ∑n=1∞n+2n2+5\displaystyle\sum_{n=1}^{\infty} \dfrac{\sqrt{n} + 2}{n^{2} + 5} converge or diverge?

  1. 1
    Keep only the dominant terms to build bnb_n

    an≈nn2=1n3/2a_n \approx \dfrac{\sqrt{n}}{n^{2}} = \dfrac{1}{n^{3/2}}, so bn=1n3/2b_n = \dfrac{1}{n^{3/2}}.

  2. 2
    Decide ∑bn\sum b_n: a pp-series or a geometric series

    p=32>1p = \tfrac{3}{2} > 1: ∑bn\sum b_n converges.

  3. 3
    Compare: compute L=lim⁡anbnL = \lim \dfrac{a_n}{b_n}, or prove an inequality

    anbn=n3/2(n+2)n2+5=n2+2n3/2n2+5→1\dfrac{a_n}{b_n} = \dfrac{n^{3/2}(\sqrt{n} + 2)}{n^{2} + 5} = \dfrac{n^{2} + 2n^{3/2}}{n^{2} + 5} \to 1.

  4. 4
    Conclude: ∑an\sum a_n does what ∑bn\sum b_n does

    L=1L = 1, so the series converges by the limit comparison test.

Answer:

Converges, by limit comparison with a pp-series.

Common mistakes4
  • Getting L=0L = 0 or L=∞L = \infty and concluding anyway. Pick a better bnb_n.
  • An inequality the wrong way proves nothing: an≥bna_n \geq b_n with ∑bn\sum b_n convergent is inconclusive.
  • Comparing n2+1\sqrt{n^{2} + 1} with n2n^{2}: under a square root the power halves, so it behaves like nn.
  • Using either test on terms that change sign. Compare ∣an∣|a_n| instead.
4

Use the ratio test on a series with factorials or powers

e.g. “Factorials in the terms: n!n!, (2n)!(2n)!, (n!)2(n!)^{2}”
  1. 1
    Write an+1a_{n+1} by replacing every nn with n+1n + 1
  2. 2
    Simplify ∣an+1an∣\left|\dfrac{a_{n+1}}{a_n}\right|, canceling factorials and powers

    (n+1)!n!=n+1\dfrac{(n+1)!}{n!} = n + 1,  (2n+2)!(2n)!=(2n+2)(2n+1)\ \dfrac{(2n+2)!}{(2n)!} = (2n+2)(2n+1),  5n+15n=5\ \dfrac{5^{n+1}}{5^{n}} = 5.

  3. 3
    Take the limit LL

    (1+1n)n→e\left(1 + \tfrac{1}{n}\right)^{n} \to e is often needed when nnn^{n} appears.

  4. 4
    L<1L < 1 converges, L>1L > 1 diverges, L=1L = 1 says nothing

    L<1L < 1 means the series . With L=1L = 1, switch tests.

Remember: Next over current; below 1 converges.

See it in a problem3 worked examples

Does ∑n=1∞n23n\displaystyle\sum_{n=1}^{\infty} \dfrac{n^{2}}{3^{n}} converge or diverge?

  1. 1
    Write an+1a_{n+1} by replacing every nn with n+1n + 1

    an+1=(n+1)23n+1a_{n+1} = \dfrac{(n+1)^{2}}{3^{n+1}}.

  2. 2
    Simplify ∣an+1an∣\left|\dfrac{a_{n+1}}{a_n}\right|, canceling factorials and powers

    an+1an=(n+1)23n+1⋅3nn2=13(n+1n)2\dfrac{a_{n+1}}{a_n} = \dfrac{(n+1)^{2}}{3^{n+1}} \cdot \dfrac{3^{n}}{n^{2}} = \dfrac{1}{3}\left(\dfrac{n+1}{n}\right)^{2}.

  3. 3
    Take the limit LL

    L=13L = \dfrac{1}{3}.

  4. 4
    L<1L < 1 converges, L>1L > 1 diverges, L=1L = 1 says nothing

    L<1L < 1: converges by the ratio test.

Answer:

Converges, by the ratio test (L=13L = \tfrac{1}{3}).

Common mistakes3
  • Writing the next factorial wrong: for (2n)!(2n)! it is (2(n+1))!=(2n+2)!(2(n+1))! = (2n + 2)!, not (2n)!+1(2n)! + 1 or (2n+1)!(2n + 1)!.
  • Using the ratio test on rational terms like n2+1n4+3\dfrac{n^{2} + 1}{n^{4} + 3}: L=1L = 1 every time. Compare instead.
  • Flipping the ratio: it is an+1an\dfrac{a_{n+1}}{a_n}, next over current.
5

Decide if a series converges absolutely or conditionally

e.g. “Determine whether the series is absolutely convergent, conditionally convergent, or divergent”
  1. 1
    Drop the sign and test ∑∣an∣\sum |a_n|

    Use a pp-series, a comparison or the ratio test.

  2. 2
    If ∑∣an∣\sum |a_n| converges: absolutely convergent, done
  3. 3
    Otherwise check that bn=∣an∣b_n = |a_n| decreases to 00

    This is the alternating series test. It only applies when the signs strictly alternate.

  4. 4
    Both hold: conditionally convergent. bn↛0b_n \not\to 0: divergent

Remember: Absolute first, then alternating.

See it in a problem4 worked examples

Is ∑n=1∞(−1)nn+3\displaystyle\sum_{n=1}^{\infty} \dfrac{(-1)^{n}}{\sqrt{n} + 3} absolutely convergent, conditionally convergent, or divergent?

  1. 1
    Drop the sign and test ∑∣an∣\sum |a_n|

    ∣an∣=1n+3|a_n| = \dfrac{1}{\sqrt{n} + 3}; limit comparison with 1n\dfrac{1}{\sqrt{n}} gives L=1L = 1, and p=12≤1p = \tfrac{1}{2} \leq 1: ∑∣an∣\sum |a_n| diverges.

  2. 2
    If ∑∣an∣\sum |a_n| converges: absolutely convergent, done

    Not absolutely convergent.

  3. 3
    Otherwise check that bn=∣an∣b_n = |a_n| decreases to 00

    bn=1n+3b_n = \dfrac{1}{\sqrt{n} + 3} decreases (the bottom grows) and tends to 00.

  4. 4
    Both hold: conditionally convergent. bn↛0b_n \not\to 0: divergent

    Converges by the alternating series test, so it is .

Answer:

Conditionally convergent.

Common mistakes4
  • Calling a series conditionally convergent without showing that ∑∣an∣\sum |a_n| diverges.
  • Using the alternating series test on sin⁡n\sin n terms: their signs do not alternate. Use absolute convergence.
  • Checking bn→0b_n \to 0 but not that bnb_n decreases, or the reverse. The test needs both.
  • (−1)2n+1=−1(-1)^{2n+1} = -1 for every nn: that series is not alternating at all.
6

Estimate the error of an alternating series

e.g. “According to the Alternating Series Estimation Theorem, what is the smallest value of NN …”
  1. 1
    Check the alternating series test: bnb_n decreases to 00

    This is why the estimate is allowed; say so for the marks.

  2. 2
    Bound the error: ∣S−SN∣≤bN+1|S - S_N| \leq b_{N+1}

    The bound is the first term you leave out.

  3. 3
    Solve bN+1≤b_{N+1} \leq tolerance for the smallest NN

    With factorials or powers, try values of NN one by one. If asked, write out SNS_N.

Remember: The error is smaller than the first term left out.

See it in a problem3 worked examples

How many terms of ∑n=1∞(−1)n+1n4\displaystyle\sum_{n=1}^{\infty} \dfrac{(-1)^{n+1}}{n^{4}} guarantee an error of at most 10−410^{-4}?

  1. 1
    Check the alternating series test: bnb_n decreases to 00

    bn=1n4b_n = \dfrac{1}{n^{4}} is positive, decreasing and tends to 00.

  2. 2
    Bound the error: ∣S−SN∣≤bN+1|S - S_N| \leq b_{N+1}

    ∣S−SN∣≤1(N+1)4|S - S_N| \leq \dfrac{1}{(N+1)^{4}}.

  3. 3
    Solve bN+1≤b_{N+1} \leq tolerance for the smallest NN

    1(N+1)4≤10−4\dfrac{1}{(N+1)^{4}} \leq 10^{-4} means (N+1)4≥104(N+1)^{4} \geq 10^{4}, so N+1≥10N + 1 \geq 10 and N=9N = 9.

Answer:

N=9N = 9 terms.

Common mistakes3
  • Using bNb_N instead of bN+1b_{N+1}: the bound is the next term, the first one not added.
  • Counting terms wrong when the sum starts at n=0n = 0: then adding NN terms leaves out bNb_N first. Count the terms you add.
  • Using this bound on a series that does not alternate, or whose bnb_n does not decrease.

Need the formulas? Must memorize · Flashcards

Found this helpful? Leave us a review on Google ↗Spotted a mistake?

Which topics are you reviewing?

The topic list, Flashcards, Must memorize and Exam words will only show these. Change it any time with the Topics button above the topic list.

The integral
Techniques of integration
Applications
Sequences & series
7 of 20 topics selected