The Divergence Test and Cauchy Criterion for Series

Does the series $\sum_{n=1}^\infty \frac{n}{n+1}$ converge? A student who only checks that the terms are positive and bounded might hesitate. But a single glance settles it: the terms tend to $1$ , not $0$ . So the sum cannot possibly settle down to a finite value. This one observation — that a convergent series must have terms shrinking to zero — is the cheapest and most frequently used test in all of analysis. It is the first thing one should check before attempting anything more elaborate. But the zero-limit condition is only a necessary condition, never sufficient. Understanding exactly why it fails to be sufficient is one of the most instructive moments in a first course in analysis. The harmonic series $\sum 1/n$ is the standard cautionary tale: its terms vanish, yet the series diverges. Explaining this requires a finer tool than "look at the terms." It requires controlling entire blocks of terms at once. That finer tool is the Cauchy criterion for series, a direct translation of the completeness of $\mathbb{R}$ into the language of partial sums. This article develops both ideas rigorously and shows how they fit together. The Divergence Test (sometimes called the $n$ -th Term Test) is a corollary of the Cauchy criterion. It is not an independent fact requiring separate machinery. Deriving it this way keeps the logical structure honest and non-circular. The Cauchy criterion itself is a precise restatement of the Cauchy Sequences and Completeness property of $\mathbb{R}$ . It applies this property to the sequence of partial sums of a series. It is the theoretical bedrock underneath essentially every convergence test one meets later, including the Comparison Test. By the end, you will be able to state and prove both results with full rigor. You will see exactly where the harmonic series exploits the gap between "necessary" and "sufficient." You will understand why these two tools are the correct starting point. This holds before moving on to sharper convergence criteria. It also applies before exploring the distinction between Absolute vs Conditional Convergence.

Key takeaways

A series $\sum a_n$ can converge only if $a_n \to 0$ ; this is the Divergence Test, used to instantly rule out convergence.
The converse is false: $a_n \to 0$ does not imply convergence — the harmonic series $\sum 1/n$ is the classic counterexample.
The Cauchy criterion says $\sum a_n$ converges iff for every $\varepsilon>0$ there is $N$ with $\left|\sum_{k=m+1}^n a_k\right|<\varepsilon$ for all $n>m\ge N$ .
The Cauchy criterion is completeness of $\mathbb{R}$ applied to partial sums; it works because Cauchy sequences of reals converge.
The Divergence Test follows from the Cauchy criterion by taking $n=m+1$ : single terms are a special case of a block of terms.
Key idea: convergence of a series is controlled not by individual terms but by tails of consecutive terms becoming uniformly small.

Where this lives today

The Cauchy criterion is the theoretical engine behind nearly every convergence test taught afterward: the Comparison Test, ratio test, root test, and integral test are all typically proved by bounding tail block sums and invoking exactly this criterion, or by reducing to a series already known to converge via comparison. The distinction it enables — between necessary zero-limit behavior and genuine tail-control — becomes essential when one studies Absolute vs Conditional Convergence, since conditionally convergent series (like the alternating harmonic series) satisfy the Divergence Test's necessary condition while requiring delicate Cauchy-type block estimates to establish convergence at all. Beyond pure analysis, the same block-sum idea underlies numerical analysis, where truncation error estimates for infinite series (used to approximate constants like $\pi$ , $e$ , or values of special functions on a computer) are literally bounds on Cauchy tail blocks, telling an engineer how many terms are needed for a desired precision. In physics, perturbation series and Fourier series convergence questions rest on the same completeness argument. In probability theory, the almost-sure convergence of random series (e.g., in the study of random walks) is frequently established via a probabilistic analogue of the Cauchy criterion, such as Kolmogorov's convergence criteria for sums of independent random variables. In functional analysis, the criterion generalizes to convergence of series in Banach spaces, where completeness of the space (not just of $\mathbb{R}$ ) is exactly the hypothesis that makes the same proof go through unchanged.

Ask the tutor: Can you show me how the Cauchy criterion is used to prove the Comparison Test?

(a_n)_{n=1}^\infty

Intuition

Think of building a tower one brick ( $a_n$ ) at a time, where $s_N$ is the height after $N$ bricks. For the tower's height to converge to a finite limiting value $L$ , the height increments must eventually become negligible — you cannot keep adding bricks of a fixed nonzero size and expect the total height to stabilize. This is the content of the Divergence Test: if $a_n \not\to 0$ , the increments never shrink, so the partial sums cannot settle down. But shrinking increments alone do not guarantee a finite tower. The harmonic series adds bricks of size $1/n$ , individually tiny, but the cumulative effect of many small bricks in a row can still push the height to infinity. The Cauchy criterion diagnoses this precisely: it asks not "is the next brick small?" but "is the total height added over any long stretch of future bricks small?" For the harmonic series, the block from $n+1$ to $2n$ always contributes more than $\frac12$ , no matter how far out you start — so the Cauchy condition fails even though individual terms vanish.

Figure

Partial sums of the harmonic series stall in growth but never stabilize

The formal results

(a_n)_{n=1}^\infty

\mathbb{R}

n

\sum_{n=1}^\infty a_n

Note the logical dependency is one-directional and non-circular: the Cauchy criterion (a biconditional resting on completeness of $\mathbb{R}$ ) is proved first, and the Divergence Test is then derived from it as a special case, rather than being independently postulated or used to help prove the Cauchy criterion.

Worked examples

Example 1 — Using the Divergence Test to instantly kill a series. Determine whether $\displaystyle\sum_{n=1}^\infty \frac{3n^2 - 5}{7n^2 + 2n - 1}$ converges. Compute the limit of the general term:

\lim_{n\to\infty} \frac{3n^2 - 5}{7n^2+2n-1} = \lim_{n\to\infty} \frac{3 - 5/n^2}{7 + 2/n - 1/n^2} = \frac{3}{7} \ne 0,

dividing numerator and denominator by $n^2$ and using that $5/n^2, 2/n, 1/n^2 \to 0$ . Since the terms tend to $3/7 \ne 0$ , the Divergence Test says the series diverges. No comparison test, integral test, or ratio test is needed — and in fact none of those tests would even apply cleanly here, since this term does not obviously compare to a standard series in a useful way. This is exactly the value of checking the Divergence Test first: it is a two-line computation that immediately settles a series that might otherwise tempt a student into a long-winded (and unnecessary) argument.

Example 2 — Why the Divergence Test alone cannot decide convergence: verifying the Cauchy criterion fails for the harmonic series. Consider $\displaystyle\sum_{n=1}^\infty \frac{1}{n}$ . Here $a_n = 1/n \to 0$ , so the Divergence Test gives no information — it is consistent with either convergence or divergence. We must use the Cauchy criterion directly. Claim: the Cauchy condition fails. It suffices to exhibit one $\varepsilon > 0$ for which no $N_0$ works, i.e., for every candidate $N_0$ we can find $n > m \ge N_0$ with the block sum $\ge \varepsilon$ . Take $\varepsilon = \tfrac12$ . Fix any $N_0 \in \mathbb{N}$ , and let $m = N_0$ (or any integer $\ge N_0$ ), $n = 2m$ . Then

\sum_{k=m+1}^{2m} \frac{1}{k} = \frac{1}{m+1} + \frac{1}{m+2} + \cdots + \frac{1}{2m}.

This sum has exactly $m$ terms, and every term is $\ge \dfrac{1}{2m}$ (since each denominator $k$ satisfies $m+1 \le k \le 2m$ , hence $\frac1k \ge \frac{1}{2m}$ ). Therefore

\sum_{k=m+1}^{2m} \frac{1}{k} \ge m \cdot \frac{1}{2m} = \frac{1}{2}.

So the block sum from $m+1$ to $2m$ is always at least $\tfrac12 = \varepsilon$ , for every choice of $m$ , no matter how large. Hence for $\varepsilon = 1/2$ , there is no $N_0$ that forces all such block sums below $\varepsilon$ : given any proposed $N_0$ , the pair $(n,m) = (2N_0, N_0)$ produces a block sum $\ge 1/2$ , violating the required bound. By the Cauchy Criterion for Series (contrapositive direction), $\sum 1/n$ diverges, despite $1/n \to 0$ . This confirms concretely that the zero-limit condition in the Divergence Test is necessary but never sufficient, and it shows the Cauchy criterion doing real diagnostic work that the Divergence Test cannot.

a_n \to 0

Scope, edge cases & common traps

Situation	What holds	What fails / must not be assumed
$a_n \not\to 0$	Series diverges (Divergence Test)	—
$a_n \to 0$	No conclusion either way	Cannot conclude convergence (harmonic series)
$a_n \to 0$ and terms alternate in sign, decreasing in absolute value	May converge (this needs a separate theorem, e.g. the Alternating Series Test)	Divergence Test alone is insufficient to prove convergence
Series converges	$a_n \to 0$ necessarily	—
Cauchy condition holds for some fixed block length only (e.g. only pairs $n=m+1$ )	Only gives $a_n\to 0$	Does not give full Cauchy condition for all $n>m$ ; this is the trap in Example 2
Complex-valued or vector-valued series in a complete normed space	Both theorems generalize verbatim, using the norm in place of $\|\cdot\|$	Completeness of the ambient space is still essential
Series over $\mathbb{Q}$ instead of $\mathbb{R}$	Cauchy criterion as an equivalence fails	$\mathbb{Q}$ is not complete: a Cauchy sequence of partial sums need not have a limit in $\mathbb{Q}$

Common trapThe classic error: treating the Divergence Test as if it could prove convergence

The single most common mistake is asserting " $a_n \to 0$ , therefore $\sum a_n$ converges," silently using the unproved converse of the Divergence Test. The Divergence Test is a one-directional implication — convergence forces $a_n \to 0$ — and its contrapositive is the only valid inferential use (nonzero-limit terms force divergence). Concluding convergence from $a_n \to 0$ alone requires an entirely different argument, such as the Cauchy criterion applied directly to the partial sums, a comparison with a known convergent series, or a specialized test (ratio, root, integral, alternating series). Always check which direction of an "if and only if is not being claimed" you are actually allowed to use.

The mathematician behind it

Origins · 1 minAugustin-Louis Cauchy and how this was discovered

Augustin-Louis Cauchy

1789–1857 · Early 19th-century French analysis, École Polytechnique

Cauchy was among the first mathematicians to insist that calculus be built on precise limit definitions rather than intuitive appeals to infinitesimals or geometric plausibility. His 1821 textbook Cours d'Analyse formalized the modern $\varepsilon$ -based definition of limits, continuity, and convergence, and it is there that the criterion bearing his name — a sequence converges if and only if its terms eventually cluster arbitrarily close together — first appears in something close to its modern form. Applied to partial sums of a series, this criterion becomes exactly the tool used throughout this article.

History

- 1821 — Cauchy publishes Cours d'Analyse, giving a rigorous

\varepsilon

-style treatment of limits and stating the Cauchy convergence criterion for sequences.

- 1830s–1850s — The criterion is gradually recognized as depending on a completeness property of the real numbers not yet formally isolated in Cauchy's own era.

- 1872 — Cantor and Méray, working independently, construct the real numbers using equivalence classes of Cauchy sequences, finally making rigorous the completeness property Cauchy's criterion silently required.

- 1872 — Dedekind constructs

\mathbb{R}

via cuts, giving an equivalent route to completeness and confirming the Cauchy criterion as a genuine theorem rather than an assumption.

- Late 19th century — The Cauchy criterion for series and the Divergence Test become standard fixtures of rigorous analysis textbooks, in the form presented here.

Is the Divergence Test the same thing as the Cauchy criterion?

No. The Cauchy criterion is a full if-and-only-if characterization of convergence, controlling arbitrary blocks $\sum_{k=m+1}^n a_k$ of consecutive terms. The Divergence Test is a one-directional consequence of it, obtained by looking only at blocks of length one (single terms $a_{m+1}$ ). The Divergence Test is weaker: it can prove divergence when terms don't vanish, but it can never prove convergence, since the converse implication ( $a_n \to 0 \Rightarrow$ convergence) is false, as the harmonic series demonstrates. The Cauchy criterion, by contrast, is strong enough to fully decide convergence in principle, though applying it in practice can require cleverness in bounding the block sums.

Can the Divergence Test ever be used to prove that a series converges?

No, never. The Divergence Test has the strict logical form "convergence implies $a_n \to 0$ ," and its only valid use is the contrapositive: if $a_n \not\to 0$ , the series diverges. Observing that $a_n \to 0$ tells you nothing about convergence — it is equally consistent with convergent series (like $\sum 1/n^2$ ) and divergent ones (like $\sum 1/n$ ). To prove convergence you need a genuinely different tool: the Cauchy criterion applied directly, a comparison to a known convergent series, the ratio or root test, the integral test, or a specialized test like the Alternating Series Test.

Why does the harmonic series diverge if its terms shrink to zero?

Individual terms shrinking to zero controls the size of each brick, but convergence of the series requires that sums of arbitrarily many consecutive future bricks also shrink to zero — a much stronger requirement. For the harmonic series, grouping terms from $n+1$ to $2n$ always yields a block sum of at least $1/2$ , regardless of how large $n$ is, because there are $n$ terms each at least $1/(2n)$ . Repeating this doubling argument shows the partial sums grow without bound (roughly like $\ln N$ ), even though each individual term is tiny. The decay rate $1/n$ is simply too slow to prevent this accumulation.

What role does completeness of the real numbers play in the Cauchy criterion?

It is the entire content of the "hard direction" of the proof. Showing that convergence implies the Cauchy block condition only needs the triangle inequality. But showing that the Cauchy block condition implies convergence requires knowing that Cauchy sequences of real numbers actually have a limit in $\mathbb{R}$ — this is precisely the completeness property, equivalent to the least-upper-bound axiom. Over a non-complete ordered field like $\mathbb{Q}$ , there exist sequences of partial sums that satisfy the Cauchy condition (their terms cluster together) yet fail to converge to any rational number, so the criterion as an equivalence would fail there.

Does the Cauchy criterion apply to series with complex or vector-valued terms?

Yes, provided the ambient space is complete. If $a_n$ lies in $\mathbb{C}$ (complete under $|\cdot|$ ) or in a Banach space (a complete normed vector space), the exact same proof goes through: convergence of partial sums is equivalent to the Cauchy condition on block sums, because the underlying completeness property still guarantees that Cauchy sequences converge. This is why the criterion generalizes so cleanly to functional analysis, where it underlies convergence tests for series of functions in spaces like $L^p$ or $C[a,b]$ .

If a series passes the Divergence Test (terms go to zero), what should I try next?

Check whether the terms are eventually positive and compare the series to a known benchmark using the Comparison Test, or check whether the series is absolutely convergent by testing $\sum |a_n|$ , connecting to the theory of Absolute vs Conditional Convergence. For series with a recognizable algebraic or exponential structure, the ratio or root test is often decisive. For alternating series with monotonically decreasing terms tending to zero, the Alternating Series Test gives convergence directly. The Divergence Test is only ever a first filter, never a final verdict when it is satisfied.

Is it true that if the partial sums are bounded, the series converges?

Not in general — boundedness of partial sums alone is not equivalent to convergence for arbitrary real series; for example $s_N$ could oscillate between two values without settling (e.g. $a_n = (-1)^n$ gives bounded but non-convergent partial sums $1,0,1,0,\dots$ , and indeed this fails the Divergence Test too, since $a_n \not\to 0$ ). Boundedness plus monotonicity of partial sums does guarantee convergence, by the Monotone Convergence Theorem, which is itself another face of completeness of $\mathbb{R}$ . But boundedness by itself is weaker than the full Cauchy condition.