Calculus › Section J · Infinite series › Topic J.4
Comparison tests
A series that is not quite a p-series is usually almost one. Comparison tests line an unknown series up against a known benchmark and let the verdict transfer, with no need to sum anything.
The idea: measure against a benchmark
The integral test gave you one family you completely understand: \(\sum \frac{1}{n^p}\) converges exactly when \(p > 1\), along with geometric series. Now put that knowledge to work on series you have never seen, such as \(\sum \frac{1}{n^3 + 2n + 1}\). Its terms are slightly smaller than \(\frac{1}{n^3}\), and the \(\frac{1}{n^3}\) series is a known convergent one. If a positive series sits underneath a convergent one, its partial sums are smaller, so they cannot run away either. That is the entire idea, and the tests in this topic are two ways of making it precise.
The comparison test
Suppose \(0 \le a_n \le b_n\) for every \(n\) (it is enough for this to hold from some point onward). Then the test works in two directions. Convergent ceiling: if \(\sum b_n\) converges, then \(\sum a_n\) converges too, because the partial sums of \(a_n\) are increasing yet trapped below the finite sum of \(\sum b_n\), and the monotone convergence theorem for sequences applies to them. Divergent floor: if \(\sum a_n\) diverges, then \(\sum b_n\) diverges too, because the partial sums of \(b_n\) stand above partial sums that run past every bound. Both versions need genuinely positive terms.
Worked example: two direct comparisons
First, does \(\sum \frac{1}{n^3 + 2n + 1}\) converge? Since \(n^3 + 2n + 1 > n^3\), each term satisfies \(0 < \frac{1}{n^3 + 2n + 1} < \frac{1}{n^3}\), and \(\sum \frac{1}{n^3}\) is a convergent p-series. Smaller than convergent: it converges. Second, does \(\sum \frac{n + 2}{n^2 + 3}\) converge? Here \(\frac{1}{n}\) is the natural floor to test: \(\frac{n+2}{n^2+3} \ge \frac{1}{n}\) holds exactly when \(n(n + 2) \ge n^2 + 3\), that is \(n \ge 2\). From the second term on, the series sits above the divergent harmonic series, so it diverges. One comparison up, one down, and both directions of the test got used.
The limit comparison test
Direct comparison demands an inequality you can actually prove, and sometimes the algebra fights back. The limit comparison test trades inequalities for a limit: if \(a_n > 0\), \(b_n > 0\) and
then \(\sum a_n\) and \(\sum b_n\) live or die together. The reading is that for large \(n\) the terms are honest multiples of each other, \(a_n \approx c\,b_n\), so their running totals behave alike. Choose \(b_n\) by keeping only the dominant terms of \(a_n\): for \(\frac{2n + 3}{n^4 + 5n + 1}\), the dominant terms are \(\frac{2n}{n^4} = \frac{2}{n^3}\), so compare with \(\frac{1}{n^3}\). Indeed \(\lim \frac{(2n+3)/(n^4+5n+1)}{1/n^3} = \lim \frac{2n^4 + 3n^3}{n^4 + 5n + 1} = 2\), a healthy finite positive number, so the series converges like \(\sum \frac{1}{n^3}\).
Worked example: when direct comparison gets stuck
Does \(\sum \sin\!\left(\frac{1}{n}\right)\) converge? The terms are positive and approach 0, so the divergence test is silent. Direct comparison tempts you with \(0 < \sin\!\left(\frac{1}{n}\right) < \frac{1}{n}\), but smaller than the divergent harmonic series is exactly the setup where the comparison test says nothing. Limit comparison cuts through:
using the substitution \(x = \frac{1}{n}\) and the famous trigonometric limit. The limit is a positive finite number, so the series shares the harmonic verdict: it diverges. The lesson is general: when the useful inequality points the unhelpful way, switch to limit comparison.
Choosing between the two tests
A short decision guide. Try the divergence test first in every problem: if the terms do not approach 0, everything ends there. Next ask what the series is almost a case of: strip to dominant terms and name a p-series or geometric series with the same growth rate. If an inequality to that benchmark writes down easily, use direct comparison. If the inequality is fiddly, or the natural comparison only lands one way, compute \(\lim \frac{a_n}{b_n}\) instead. If that limit comes out 0 or \(\infty\), the test is silent and you need a different benchmark, one whose growth rate genuinely matches.
Geometric benchmarks too
p-series are not the only benchmarks; geometric series work the same way. Take \(\sum \frac{1}{2^n + n}\). Since \(2^n + n > 2^n\), every term satisfies \(0 < \frac{1}{2^n + n} < \frac{1}{2^n} = \left(\frac{1}{2}\right)^n\), which is geometric with ratio \(\frac{1}{2}\) and converges. Convergent ceiling: done. When the ratio is hidden, limit comparison finds it: \(\sum \frac{3^n}{5^n + 100}\) has dominant terms \(\left(\frac{3}{5}\right)^n\), and \(\lim \frac{3^n/(5^n + 100)}{(3/5)^n} = \lim \frac{5^n}{5^n + 100} = 1\), so it converges like a geometric series with ratio \(\frac{3}{5}\). Any series whose growth rate you can name is decidable.
Common mistakes to avoid
The error that ruins the most answers is running the inequality in the wrong direction: being smaller than a divergent series, like \(\sum \frac{1}{2n + 7}\) under \(\sum \frac{1}{n}\), proves nothing at all, and neither does being larger than a convergent one. Second, a limit comparison that returns \(c = 0\) or \(c = \infty\) is not a conclusion; it is a prompt to find a better \(b_n\). Third, both tests assume positive terms, so they cannot be applied blindly to a series with negative terms. Fourth, do not choose the benchmark by the whole expression: for \(\frac{n + 2}{n^2 + 3}\), the dominant terms give \(\frac{1}{n}\), not \(\frac{1}{n^2}\), and picking the wrong power makes the limit come out 0 or \(\infty\) and wastes the attempt.
Key takeaways
Direct comparison: \(0 \le a_n \le b_n\) with \(\sum b_n\) convergent forces \(\sum a_n\) convergent; \(\sum a_n\) divergent forces \(\sum b_n\) divergent. Smaller than divergent and larger than convergent prove nothing. Limit comparison: if \(\lim \frac{a_n}{b_n} = c\) with \(0 < c < \infty\), the two series share a verdict; pick \(b_n\) from the dominant terms. When direct comparison is awkward, limit comparison usually settles the series in one limit computation. The divergence test always goes first.
Flashcards
Flashcard
0 ≤ aₙ ≤ bₙ and Σ bₙ converges ⇒ Σ aₙ converges
Test your understanding
Question 1 of 5
Does the series 1/(n³ + 2n + 1) from n = 1 to infinity converge?
Question 2 of 5
Using limit comparison with bₙ = 1/n³, what is lim aₙ/bₙ for aₙ = (2n + 3)/(n⁴ + 5n + 1)?
Question 3 of 5
Why does Σ sin(1/n) diverge?
Question 4 of 5
A student argues: 1/(2n + 7) ≤ 1/n and Σ 1/n diverges, so Σ 1/(2n + 7) converges. What is wrong?
Question 5 of 5
Does the series (n + 2)/(n² + 3) from n = 1 to infinity converge?
Standards: 10 (BC)
