The Comparison Test

Does the series ∑1n2+n\sum \frac{1}{n^2 + \sqrt{n}} converge? Nothing we know from geometric series or telescoping tricks applies directly. Yet the terms so obviously resemble 1n2\frac{1}{n^2} that a definitive answer feels within reach. The Comparison Test turns this intuition into a rigorous method. If a series of nonnegative terms is trapped underneath a known convergent series, it too must converge. If it towers over a known divergent series, it too must diverge. This is the workhorse test of the entire theory of series. It converts the impossibly hard problem of "does this sequence of partial sums converge" into an easier one. That easier problem is: "is this series smaller or bigger than one I already understand." The idea matters because most series encountered in practice are not geometric, not telescoping. They also have no closed form for their partial sums. What we typically can see is their asymptotic size — how the terms behave for large nn. The Comparison Test, together with its refinement the Limit Comparison Test, lets us read off convergence behavior directly from this asymptotic size. To do this, we use a small stable of reference series as yardsticks (geometric series ∑rn\sum r^n, and later the pp-series ∑1/np\sum 1/n^p). The big picture is monotonicity. Partial sums of a nonnegative series form a monotone increasing sequence. So by the Monotone Convergence Theorem, such a sequence converges if and only if it is bounded above. Comparison is nothing more than an observation about monotone sequences. One bounded (or unbounded) monotone sequence can certify boundedness (or unboundedness) of another, term by term. This single idea, dressed up in two forms — direct and limit comparison — underlies essentially every convergence test for positive series that follows. This includes the Ratio Test, the Root Test, and the Integral Test. This article gives the precise statements and complete proofs of the Direct Comparison Test and the Limit Comparison Test. It builds the intuition with explicit numerical pictures, and works two nontrivial examples. Finally, it shows exactly where the nonnegativity hypothesis is essential.

Key takeaways
  • The Comparison Test applies to series with nonnegative terms and works by trapping one series between bounds set by another.
  • Direct comparison: if 0≤an≤bn0 \le a_n \le b_n eventually and ∑bn\sum b_n converges, then ∑an\sum a_n converges; if ∑an\sum a_n diverges, so does ∑bn\sum b_n.
  • The proof rests entirely on the Monotone Convergence Theorem applied to partial sums, plus the Cauchy Criterion for Series.
  • Limit comparison replaces a term-by-term inequality with an asymptotic ratio test: if an/bn→L∈(0,∞)a_n/b_n \to L \in (0,\infty), the two series share the same convergence behavior.
  • Nonnegativity cannot be dropped: alternating or sign-changing sequences can violate every conclusion of the test.
  • Key idea: convergence of nonnegative series is really about boundedness of partial sums, and boundedness transfers along inequalities.

Where this lives today

The Comparison Test is the load-bearing wall underneath nearly every other convergence test for positive series: the Ratio Test and Root Test are typically proved by comparing the series in question to a geometric series once the ratio or root condition is established, and the Integral Test compares a series to an integral by sandwiching rectangles between curves — a direct geometric analogue of term-by-term comparison. In probability theory, comparison arguments justify convergence of expectations and variances defined as infinite sums, such as verifying that a discrete random variable has finite expectation by comparing its tail probabilities to a known summable sequence (a technique closely related to the Borel–Cantelli lemmas). In numerical analysis, comparison bounds are used to estimate truncation error when approximating a series by a finite partial sum, since a convergent comparison series with known tail bounds gives explicit error control. In cryptography and algorithm analysis, asymptotic comparison of growth rates (the same an/bn→La_n/b_n \to L idea) underlies Big-OO notation and the classification of algorithmic complexity classes. In physics, perturbation series and partition function expansions are routinely tested for convergence by comparing term magnitudes to geometric or power-law benchmarks, especially in statistical mechanics and quantum field theory where exact summation is impossible but asymptotic comparison determines whether an expansion makes sense at all.

Ask the tutor: How does the Comparison Test get used to prove the Ratio Test actually works?

DefinitionComparison of nonnegative series

Let (an)n=1∞(a_n)_{n=1}^\infty and (bn)n=1∞(b_n)_{n=1}^\infty be sequences of real numbers.

Direct comparison hypothesis. Suppose there exists N∈NN \in \mathbb{N} such that for all n≥Nn \ge N,

0≤an≤bn.0 \le a_n \le b_n.

Limit comparison hypothesis. Suppose an≥0a_n \ge 0 and bn>0b_n > 0 for all sufficiently large nn, and the limit

L=lim⁡n→∞anbnL = \lim_{n\to\infty} \frac{a_n}{b_n}

exists in [0,∞][0,\infty]. In either setting we compare the two series ∑n=1∞an\sum_{n=1}^\infty a_n and ∑n=1∞bn\sum_{n=1}^\infty b_n, meaning we compare the behavior (convergence or divergence) of their sequences of partial sums

SN=∑n=1Nan,TN=∑n=1Nbn.S_N = \sum_{n=1}^N a_n, \qquad T_N = \sum_{n=1}^N b_n.

Intuition

Think of the partial sums SN=a1+a2+⋯+aNS_N = a_1 + a_2 + \cdots + a_N of a nonnegative series as a person walking only forward (or standing still) along a number line, one step an≥0a_n \ge 0 at a time. Such a walk either marches off to infinity, or it is confined to some bounded stretch of the line and, being monotone, must settle down and converge to a limit — this is precisely the Monotone Convergence Theorem in disguise. Now imagine a second walker taking steps bn≥anb_n \ge a_n, always at least as large. This walker is always at least as far along as the first. If the second walker's journey is confined to a bounded region, the first walker — being perpetually behind — must be confined too. Conversely, if the first walker's journey diverges to infinity, the second walker, always ahead, must diverge as well. This is the entire content of the Direct Comparison Test: an inequality between step sizes propagates to an inequality between the walkers' entire trajectories, and boundedness is what decides convergence for nonnegative series. The Limit Comparison Test refines this picture by weakening "always ahead" to "asymptotically proportional." If an/bn→La_n/b_n \to L with 0<L<∞0 < L < \infty, then for large nn the two walkers' step sizes are within a constant factor of each other in both directions, so neither can diverge without the other, nor converge without the other.

Figure
Trapping a series between a convergent majorant and a divergent minorant
Partial sums: comparison traps S_N between bounds N (index) sum Σ b_n → finite (e.g. Σ1/n²) Σ a_n, trapped 0≤a_n≤b_n ceiling: limit of Σ b_n Since 0 ≤ a_n ≤ b_n, the red curve (a_n's partial sums) never rises above the green one.
TheoremDirect Comparison Test

Let (an)(a_n) and (bn)(b_n) be real sequences and suppose there exists N0∈NN_0 \in \mathbb{N} such that

0≤an≤bnfor all n≥N0.0 \le a_n \le b_n \quad \text{for all } n \ge N_0.

Then: 1. If ∑n=1∞bn\sum_{n=1}^\infty b_n converges, then ∑n=1∞an\sum_{n=1}^\infty a_n converges. 2. If ∑n=1∞an\sum_{n=1}^\infty a_n diverges, then ∑n=1∞bn\sum_{n=1}^\infty b_n diverges.

Proof

Technique: monotone bounded sequences, via the Monotone Convergence Theorem, plus the tail-invariance of convergence. First note that convergence of a series is unaffected by altering finitely many terms — only the eventual behavior of partial sums matters, since two partial-sum sequences that agree in their tail differ by a fixed additive constant for all large NN. Hence it suffices to prove the theorem assuming N0=1N_0 = 1, i.e. 0≤an≤bn0 \le a_n \le b_n for all n≥1n \ge 1; the general case follows by discarding the first N0−1N_0 - 1 terms of each series, which changes convergence status of neither.

Part 1. Suppose ∑n=1∞bn\sum_{n=1}^\infty b_n converges to some real number TT. Define partial sums

SN=∑n=1Nan,TN=∑n=1Nbn.S_N = \sum_{n=1}^N a_n, \qquad T_N = \sum_{n=1}^N b_n.

Step 1: (SN)(S_N) is monotone increasing. For every NN,

SN+1−SN=aN+1≥0,S_{N+1} - S_N = a_{N+1} \ge 0,

since aN+1≥0a_{N+1} \ge 0 by hypothesis. Hence SN+1≥SNS_{N+1} \ge S_N for all NN, so (SN)(S_N) is monotone nondecreasing. Step 2: (SN)(S_N) is bounded above. Since bn≥0b_n \ge 0 for all nn (as an≤bna_n \le b_n and an≥0a_n \ge 0 forces bn≥0b_n \ge 0), the sequence (TN)(T_N) is itself monotone nondecreasing, and it converges to TT by hypothesis. A monotone nondecreasing sequence that converges must converge to its supremum (if it exceeded any term TNT_N from below infinitely often while increasing past TT that would contradict convergence), so in particular TN≤TT_N \le T for every NN. Moreover, term by term,

SN=∑n=1Nan≤∑n=1Nbn=TN≤T,S_N = \sum_{n=1}^N a_n \le \sum_{n=1}^N b_n = T_N \le T,

using an≤bna_n \le b_n for each n≤Nn \le N and summing the finitely many inequalities. Thus SN≤TS_N \le T for every NN, so (SN)(S_N) is bounded above by TT. Step 3: conclude via Monotone Convergence. By Steps 1 and 2, (SN)(S_N) is a monotone nondecreasing sequence of reals bounded above. By the Monotone Convergence Theorem, (SN)(S_N) converges to a real limit (namely sup⁡NSN\sup_N S_N, which exists and is finite since the set {SN}\{S_N\} is nonempty and bounded above). Hence ∑n=1∞an\sum_{n=1}^\infty a_n converges.

Part 2. This is the contrapositive of a companion statement, proved directly rather than assumed. Suppose ∑n=1∞an\sum_{n=1}^\infty a_n diverges. Since an≥0a_n \ge 0, the partial sums (SN)(S_N) are monotone nondecreasing (as in Step 1 above), and a monotone nondecreasing sequence either converges or diverges to +∞+\infty — it cannot oscillate, since it never decreases, so failure to converge (equivalently, failure to be bounded above, by Monotone Convergence) forces SN→∞S_N \to \infty. Since an≤bna_n \le b_n for every nn, summing gives SN≤TNS_N \le T_N for every NN, exactly as in Step 2. Because SN→∞S_N \to \infty, for every M>0M > 0 there exists N1N_1 such that SN>MS_N > M for all N≥N1N \ge N_1; then also TN≥SN>MT_N \ge S_N > M for all N≥N1N \ge N_1. Hence TN→∞T_N \to \infty as well, i.e. (TN)(T_N) diverges (to infinity), so ∑n=1∞bn\sum_{n=1}^\infty b_n diverges. This completes the proof of both parts. ■\blacksquare

∎
CorollaryContrapositive restatement

Under the hypotheses 0≤an≤bn0 \le a_n \le b_n eventually: if ∑bn\sum b_n diverges this says nothing about ∑an\sum a_n (it may converge or diverge), and if ∑an\sum a_n converges this says nothing about ∑bn\sum b_n. Only the two directions proved above — convergence transfers downward, divergence transfers upward — are valid.

TheoremLimit Comparison Test

Let (an)(a_n) and (bn)(b_n) be sequences with an≥0a_n \ge 0 and bn>0b_n > 0 for all nn sufficiently large, and suppose

L=lim⁡n→∞anbnL = \lim_{n\to\infty} \frac{a_n}{b_n}

exists in [0,∞][0, \infty]. Then: 1. If 0<L<∞0 < L < \infty, then ∑an\sum a_n converges if and only if ∑bn\sum b_n converges. 2. If L=0L = 0 and ∑bn\sum b_n converges, then ∑an\sum a_n converges. 3. If L=∞L = \infty and ∑bn\sum b_n diverges, then ∑an\sum a_n diverges.

Proof

Technique: reduce to the Direct Comparison Test using the definition of limit to extract an eventual two-sided inequality. As before, convergence is unaffected by finitely many terms, so we may assume an≥0a_n \ge 0 and bn>0b_n > 0 for all n≥1n \ge 1.

Part 1: 0<L<∞0 < L < \infty. Apply the definition of limit with ε=L/2>0\varepsilon = L/2 > 0: there exists N0N_0 such that for all n≥N0n \ge N_0,

∣anbn−L∣<L2,\left| \frac{a_n}{b_n} - L \right| < \frac{L}{2},

which rearranges to

L2 bn<an<3L2 bnfor all n≥N0.\frac{L}{2} \, b_n < a_n < \frac{3L}{2}\, b_n \qquad \text{for all } n \ge N_0.

(Here we used bn>0b_n > 0 to multiply the inequality through without reversing it.) (⇐\Leftarrow) Suppose ∑bn\sum b_n converges. Then ∑3L2bn\sum \frac{3L}{2} b_n converges too, being a constant multiple of a convergent series (this follows from linearity of limits applied to partial sums: if TN→TT_N \to T then 3L2TN→3L2T\frac{3L}{2} T_N \to \frac{3L}{2}T). Since 0≤an<3L2bn0 \le a_n < \frac{3L}{2} b_n for n≥N0n \ge N_0, the Direct Comparison Test (Part 1) gives that ∑an\sum a_n converges. (⇒\Rightarrow) Suppose ∑an\sum a_n converges. From L2bn<an\frac{L}{2} b_n < a_n for n≥N0n \ge N_0, i.e. 0≤bn<2Lan0 \le b_n < \frac{2}{L} a_n, and since ∑2Lan\sum \frac{2}{L} a_n converges (constant multiple of convergent series, L≠0L \ne 0 makes 2/L2/L well-defined), the Direct Comparison Test (Part 1) gives that ∑bn\sum b_n converges. Combining both directions: ∑an\sum a_n converges if and only if ∑bn\sum b_n converges.

Part 2: L=0L = 0. Apply the definition of limit with ε=1\varepsilon = 1: there exists N0N_0 such that for all n≥N0n \ge N_0,

∣anbn−0∣<1  ⟹  an<bn,\left| \frac{a_n}{b_n} - 0 \right| < 1 \implies a_n < b_n,

again using bn>0b_n > 0. So 0≤an<bn0 \le a_n < b_n for all n≥N0n \ge N_0. If ∑bn\sum b_n converges, the Direct Comparison Test (Part 1) directly gives that ∑an\sum a_n converges.

Part 3: L=∞L = \infty. By definition of a limit equal to +∞+\infty, for every M>0M > 0 there exists N0N_0 such that an/bn>Ma_n/b_n > M for all n≥N0n \ge N_0. Take M=1M = 1: there is N0N_0 with an/bn>1a_n/b_n > 1, i.e. an>bn≥0a_n > b_n \ge 0, for all n≥N0n \ge N_0. If ∑bn\sum b_n diverges, the Direct Comparison Test (Part 2), applied with the roles 0≤bn≤an0 \le b_n \le a_n, gives that ∑an\sum a_n diverges. ■\blacksquare

∎

Worked examples

Example 1 (Direct comparison, asymmetric bound). Determine whether

∑n=1∞3n+5n3−2n+7\sum_{n=1}^\infty \frac{3n+5}{n^3 - 2n + 7}

converges. First, note the terms are eventually nonnegative: for n≥2n \ge 2, n3−2n+7=n(n2−2)+7>0n^3 - 2n + 7 = n(n^2-2) + 7 > 0 (indeed n2≥4>2n^2 \ge 4 > 2 for n≥2n\ge 2), and the numerator 3n+5>03n+5>0 always, so an:=3n+5n3−2n+7>0a_n := \frac{3n+5}{n^3-2n+7} > 0 for n≥2n \ge 2. We seek a comparison series whose partial sums we already understand. The dominant behavior for large nn is numerator ∼3n\sim 3n, denominator ∼n3\sim n^3, suggesting comparison to ∑1n2\sum \frac{1}{n^2} — though we have not yet proved pp-series convergence in this article, we may instead compare to a series built from known convergent geometric-type control, or simply bound crudely and cite ∑1/n2\sum 1/n^2 as a known convergent benchmark (its convergence is standard and will be reproved via the Integral Test elsewhere; here we take it as an established comparison series, exactly as one uses ∑rn\sum r^n). Concretely, for n≥3n \ge 3: since n3−2n+7≥n3−2n≥n3−23n3=13n3n^3 - 2n + 7 \ge n^3 - 2n \ge n^3 - \frac{2}{3}n^3 = \frac{1}{3}n^3 (using 2n≤23n32n \le \frac{2}{3}n^3 for n≥2n \ge 2, i.e. n2≥3n^2 \ge 3, true for n≥2n\ge 2), and 3n+5≤3n+5n=8n3n + 5 \le 3n + 5n = 8n for n≥1n \ge 1 (crude but valid, since 5≤5n5 \le 5n), we get for n≥3n \ge 3:

an=3n+5n3−2n+7≤8n13n3=24n2.a_n = \frac{3n+5}{n^3-2n+7} \le \frac{8n}{\frac{1}{3}n^3} = \frac{24}{n^2}.

Since ∑1n2\sum \frac{1}{n^2} converges, so does ∑24n2=24∑1n2\sum \frac{24}{n^2} = 24\sum \frac1{n^2} (constant multiple of a convergent series). By the Direct Comparison Test, 0≤an≤24n20 \le a_n \le \frac{24}{n^2} for n≥3n \ge 3 implies

∑n=1∞3n+5n3−2n+7converges.\sum_{n=1}^\infty \frac{3n+5}{n^3-2n+7} \quad \text{converges.}

Example 2 (Limit comparison, non-obvious matching series). Determine whether

∑n=2∞n+1−n−1n\sum_{n=2}^\infty \frac{\sqrt{n+1} - \sqrt{n-1}}{n}

converges. The terms are positive: n+1>n−1\sqrt{n+1} > \sqrt{n-1} for n≥2n \ge 2, so an:=n+1−n−1n>0a_n := \frac{\sqrt{n+1}-\sqrt{n-1}}{n} > 0. The numerator is a difference of nearby square roots — exactly the shape that benefits from rationalizing:

n+1−n−1=(n+1)−(n−1)n+1+n−1=2n+1+n−1.\sqrt{n+1} - \sqrt{n-1} = \frac{(n+1)-(n-1)}{\sqrt{n+1}+\sqrt{n-1}} = \frac{2}{\sqrt{n+1}+\sqrt{n-1}}.

So

an=2n(n+1+n−1).a_n = \frac{2}{n\left(\sqrt{n+1}+\sqrt{n-1}\right)}.

For large nn, n+1+n−1∼2n\sqrt{n+1}+\sqrt{n-1} \sim 2\sqrt{n}, so an∼2n⋅2n=1n3/2a_n \sim \frac{2}{n \cdot 2\sqrt n} = \frac{1}{n^{3/2}}. This suggests comparison with bn=1n3/2b_n = \frac{1}{n^{3/2}}, a convergent pp-series (p=3/2>1p = 3/2 > 1, again taken here as an established benchmark). Compute the limit:

anbn=2n3/2n(n+1+n−1)=2nn+1+n−1.\frac{a_n}{b_n} = \frac{2 n^{3/2}}{n\left(\sqrt{n+1}+\sqrt{n-1}\right)} = \frac{2\sqrt n}{\sqrt{n+1}+\sqrt{n-1}}.

Divide numerator and denominator by n\sqrt n:

anbn=21+1/n+1−1/n→n→∞21+1=1.\frac{a_n}{b_n} = \frac{2}{\sqrt{1 + 1/n} + \sqrt{1 - 1/n}} \xrightarrow[n\to\infty]{} \frac{2}{1 + 1} = 1.

So L=1∈(0,∞)L = 1 \in (0,\infty). By the Limit Comparison Test, since ∑1n3/2\sum \frac{1}{n^{3/2}} converges, so does ∑an\sum a_n:

∑n=2∞n+1−n−1nconverges.\sum_{n=2}^\infty \frac{\sqrt{n+1}-\sqrt{n-1}}{n} \quad \text{converges.}

Notice how much harder Example 2 would be with direct comparison alone: there is no clean, easily-verified inequality an≤C/n3/2a_n \le C/n^{3/2} without first doing exactly this rationalization and limit computation — limit comparison packages that asymptotic matching into a single limit check.

CounterexampleWhy nonnegativity cannot be dropped

Consider dropping the hypothesis an≥0a_n \ge 0 (equivalently bn≥0b_n \ge 0), keeping only an≤bna_n \le b_n for all nn, and ask whether "∑bn\sum b_n converges implies ∑an\sum a_n converges" still holds. Take

an=(−1)n,bn=1if n even,bn=−12if n odd.a_n = (-1)^n, \qquad b_n = 1 \quad \text{if } n \text{ even}, \qquad b_n = -\tfrac{1}{2} \quad \text{if } n \text{ odd}.

Wait — check an≤bna_n \le b_n: for nn even, an=1≤bn=1a_n = 1 \le b_n = 1, fine; for nn odd, an=−1≤bn=−1/2a_n = -1 \le b_n = -1/2, fine. So an≤bna_n \le b_n for all nn. But ∑bn\sum b_n has partial sums oscillating and unbounded in absolute structure (it doesn't even satisfy the Divergence Test since bn↛0b_n \not\to 0), so this particular choice is not illuminating as a convergent majorant. Let's instead build a genuinely convergent majorant with a sign-changing minorant, which is the sharper counterexample: Let bn=1n2b_n = \frac{1}{n^2} for all n≥1n \ge 1 (so ∑bn\sum b_n converges), and let

an=−1+1n2.a_n = -1 + \frac{1}{n^2}.

Then an≤bna_n \le b_n holds for all nn (since −1≤0-1 \le 0). But an→−1≠0a_n \to -1 \ne 0, so by the Divergence Test, ∑an\sum a_n diverges — indeed SN=∑n=1Nan→−∞S_N = \sum_{n=1}^N a_n \to -\infty. This shows: without an≥0a_n \ge 0, an upper bound an≤bna_n \le b_n by a convergent series ∑bn\sum b_n gives no control at all on convergence of ∑an\sum a_n, because nothing prevents ana_n from being very negative. The entire mechanism of the proof — monotonicity of partial sums bounding them above by the limit of (TN)(T_N) — required an≥0a_n \ge 0 to get monotonicity of (SN)(S_N) in the first place; without it, (SN)(S_N) need not be monotone and boundedness-from-above says nothing about convergence (a bounded-above but not bounded-below, non-monotone sequence can still diverge, as here SN→−∞S_N \to -\infty is in fact unbounded below). This is exactly why the theorem's hypothesis specifies nonnegative terms, not just an inequality between them.

Scope, edge cases & common traps

SituationWhat to checkCommon mistake
Terms not eventually nonnegativeRewrite using ∣an∣|a_n| and test Absolute Convergence insteadApplying comparison directly to signed series
Only finitely many terms violate an≤bna_n \le b_nStill fine — convergence only depends on the tailConcluding the test fails because of a few early exceptions
L=0L = 0 in limit comparisonOnly get one direction: ∑bn\sum b_n converges ⇒∑an\Rightarrow \sum a_n convergesAssuming L=0L=0 gives an "if and only if" — it does not
L=∞L = \infty in limit comparisonOnly get: ∑bn\sum b_n diverges ⇒∑an\Rightarrow \sum a_n divergesAssuming L=∞L=\infty says anything when ∑bn\sum b_n converges
Choosing comparison seriesMatch asymptotic order (degree, exponential rate)Comparing to a series with the wrong growth rate, making the inequality false
Both series divergeDirect comparison never proves divergence of the smaller oneTrying to conclude ∑an\sum a_n diverges from ∑bn\sum b_n diverges when an≤bna_n \le b_n

The mathematician behind it

Origins · 1 minAugustin-Louis Cauchy and how this was discovered
AC
Augustin-Louis Cauchy
1789–1857 · Founder of rigorous analysis in early 19th-century Paris

Cauchy was the central figure in transforming calculus from a computational art, inherited from Newton and Leibniz, into a rigorous logical structure built on precise definitions of limit, continuity, and convergence. Working at the École Polytechnique in the 1820s, he formulated the modern ε\varepsilon-based definition of limit and used it to give the first fully rigorous treatments of series convergence, including comparison-style arguments that trap an unknown series between known bounds. His 1821 Cours d'Analyse is the direct ancestor of every modern real analysis textbook's treatment of series, including the comparison and ratio tests that bear methodological echoes of his work throughout this article.

History
- 1821 — Cauchy's Cours d'Analyse gives the first rigorous ε\varepsilon-style definitions of limit and convergence of series, laying the logical groundwork for comparison-based arguments.
- 1826 — Cauchy and later Niels Henrik Abel study convergence criteria for series systematically, including precursors to ratio and comparison ideas.
- 1830s–1840s — Comparison-style reasoning becomes standard technique alongside the emerging Ratio Test and the study of pp-series, cementing ∑1/np\sum 1/n^p and ∑rn\sum r^n as the canonical comparison benchmarks.
- 1893 — Camille Jordan's Cours d'analyse and similar textbooks formalize the Direct and Limit Comparison Tests essentially in the form taught today.
- 20th century — The Comparison Test becomes a foundational lemma from which the Ratio Test, Root Test, and Integral Test are all derived, cementing its role as the base case of positive-series convergence theory.
Why must the terms be nonnegative for the Comparison Test?

Nonnegativity is what makes the sequence of partial sums monotone nondecreasing: SN+1−SN=aN+1≥0S_{N+1} - S_N = a_{N+1} \ge 0. Monotone sequences have a very special property — by the Monotone Convergence Theorem, a monotone nondecreasing sequence converges if and only if it is bounded above, with no other possible behavior (no oscillation). Comparison is fundamentally an argument about transferring an upper bound from one monotone sequence to another. Without nonnegativity, partial sums can rise and fall, so being bounded above (or below) no longer determines convergence, and an inequality an≤bna_n \le b_n between terms says nothing reliable about the relationship between the two partial-sum sequences' limits.

Can I use the Comparison Test on series with negative terms?

Not directly, but indirectly yes: apply the Comparison Test to ∣an∣|a_n| instead of ana_n. If ∣an∣≤bn|a_n| \le b_n eventually and ∑bn\sum b_n converges, then ∑∣an∣\sum |a_n| converges, which means ∑an\sum a_n converges absolutely and therefore converges (absolute convergence implies convergence, a separate theorem proved via the Cauchy Criterion). This is how the Comparison Test extends its reach to signed series in practice — always through the intermediary of absolute values, never applied to the signed terms themselves.

How do I choose the right comparison series?

Identify the dominant term controlling the growth of ana_n as n→∞n \to \infty: look at the highest powers of nn in a rational expression, the dominant exponential in an exponential expression, or the leading behavior after simplifying roots and logarithms. Build bnb_n from that dominant behavior alone, discarding lower-order corrections. Then verify either a direct inequality an≤C⋅bna_n \le C \cdot b_n for a suitable constant CC (Direct Comparison), or compute lim⁡an/bn\lim a_n/b_n and check it is a positive finite number (Limit Comparison). The Limit Comparison Test is usually easier to execute because it avoids the algebraic work of proving an exact inequality.

What is the difference in strength between Direct and Limit Comparison?

They are logically equivalent in the sense that each can be derived from the other combined with basic limit facts, but they differ in ease of use. Direct comparison demands you exhibit an actual inequality an≤bna_n \le b_n (or an≥bna_n \ge b_n) valid for all sufficiently large nn, which can require delicate algebraic manipulation. Limit comparison only requires computing a single limit of a ratio, which is often a routine application of standard limit laws. In practice, limit comparison is preferred whenever the terms involve additions, subtractions, or roots that make a clean inequality awkward to produce, as in the rationalization example above.

Does the Comparison Test tell me the value of a convergent series?

No. The Comparison Test only certifies whether a series converges or diverges; it gives no information about the numerical value of the sum. To find the actual sum, one typically needs a closed-form technique like a geometric or telescoping structure, or a specialized method (Fourier analysis, generating functions, residue calculus) depending on the series. The Comparison Test's value is purely as a convergence-detection tool, analogous to how the Divergence Test only detects divergence and gives no information when it is inconclusive.

Is the Comparison Test the same as the Squeeze Theorem for sequences?

They share the same spirit — trapping an unknown quantity between two known bounds — but they are not identical. The Squeeze Theorem for sequences concludes that a sequence converges to a specific limit LL when trapped between two sequences both converging to LL. The Comparison Test for series concludes only convergence or divergence (a yes/no boundedness question), typically using a one-sided bound (only an upper bound, or only a lower bound), and does not require knowing or matching an exact limiting value. The underlying tool for both, however, is the same idea of controlling one quantity via known bounds on another.

What comes next after mastering the Comparison Test?

The Comparison Test is the foundation for essentially all subsequent convergence tests for positive series. The Ratio Test and Root Test examine the asymptotic ratio an+1/ana_{n+1}/a_n or the value ann\sqrt[n]{a_n}, and their proofs work by comparing the tail of the series to a geometric series once that ratio or root is shown to be eventually less than some r<1r < 1. The Integral Test compares a series to an integral, using monotonicity of the summand to sandwich the partial sums between two integral bounds. Understanding comparison deeply — especially why nonnegativity and monotonicity of partial sums are the load-bearing hypotheses — makes all of these later tests feel like natural extensions rather than new tricks.

Before this

Where this leads