Skip to content
Topos Uranos

Alternating series and conditional convergence

Summary

The lesson on numerical series and their convergence proved the Cauchy criterion for series and that the harmonic series diverges, even though its general term tends to 00; the lessons that followed gave tests for series with nonnegative terms. This lesson studies the series that converge because their terms, of varying signs, cancel each other out. If the series of absolute values converges, the series converges, and it is said to converge absolutely; if it converges while the series of absolute values does not, it converges conditionally, like the integrals of the lesson on the conditional convergence of improper integrals. The Leibniz test decides the convergence of the alternating series whose terms decrease in absolute value toward 00, and it also gives the sign of the remainder and bounds it, in absolute value, by the first omitted term; with it, the alternating harmonic series converges conditionally. Abel’s summation by parts, a discrete version of integration by parts, proves Dirichlet’s and Abel’s tests for series, parallel to those for integrals, and with them the series of sin⁡(nx)n\frac{\sin(nx)}{n} and of cos⁡(nx)n\frac{\cos(nx)}{n} converge when xx is not a multiple of 2π2\pi.

Learning objectives

  1. Distinguish absolute from conditional convergence, and prove that every absolutely convergent series converges.
  2. Prove the Leibniz test with the sign and the bound of the remainder, and use it to approximate the sum of an alternating series with a guaranteed error.
  3. Prove Abel’s summation by parts and, with it, Dirichlet’s and Abel’s tests for series, and recognize which hypotheses cannot be omitted.
  4. Decide whether a series of varying signs, such as ∑n=1∞sin⁡(nx)n\sum_{n=1}^{\infty} \frac{\sin(nx)}{n}, converges absolutely or conditionally.

Absolute and conditional convergence

The tests of the preceding lessons, such as the theorems of the comparison test for series and of the limit comparison test for series, from the lesson on series with positive terms, require nonnegative terms. For a series of varying signs, applied to the series of absolute values, they decide whether the latter converges; in general, it remains to know what this says about the series itself, and what happens when the series of absolute values diverges. The lesson on conditional convergence of improper integrals settled both questions for integrals, and this lesson does so for series, with the conventions of the lesson on numerical series and their convergence: partial sums SnS_n, sum SS and remainder Rn=S−SnR_n = S - S_n.

DefinitionAbsolute and conditional convergence

Let (an)(a_n) be a sequence of real numbers. The series ∑n=1∞an\sum_{n=1}^{\infty} a_n converges absolutely if the series ∑n=1∞∣an∣\sum_{n=1}^{\infty} \abs{a_n} converges, and converges conditionally if it converges and the series ∑n=1∞∣an∣\sum_{n=1}^{\infty} \abs{a_n} diverges. Likewise for a series starting from an index n0n_0.

TheoremAn absolutely convergent series converges

If the series ∑n=1∞∣an∣\sum_{n=1}^{\infty} \abs{a_n} converges, the series ∑n=1∞an\sum_{n=1}^{\infty} a_n also converges, and its sum satisfies the following bound:

∣∑n=1∞an∣≤∑n=1∞∣an∣\abs{\sum_{n=1}^{\infty} a_n} \leq \sum_{n=1}^{\infty} \abs{a_n}

The sum of a stretch of terms does not exceed in absolute value the sum of their absolute values, and the Cauchy criterion passes from one series to the other.

Proof

  1. m>n≥N→∑k=n+1m∣ak∣<ε\dato{cau}{m > n \geq N \rightarrow \sum_{k=n + 1}^{m} \abs{a_k} < \varepsilon}

    Given ε>0\varepsilon > 0, the Cauchy criterion for series, from the lesson on numerical series and their convergence, applied to the series of absolute values, gives a natural number NN, depending only on ε\varepsilon, such that the sum of every stretch after NN has absolute value less than ε\varepsilon; since its terms are not negative, that sum is its own absolute value.

  2. m>n≥N→∣∑k=n+1mak∣≤∑k=n+1m∣ak∣<εm > n \geq N \rightarrow \abs{\sum_{k=n + 1}^{m} a_k} \leq \sum_{k=n + 1}^{m} \abs{a_k} < \resaltar{\varepsilon}

    The triangle inequality, from the lesson on the real line, extended by induction to any number of summands, bounds the sum of each stretch of the series ∑n=1∞an\sum_{n=1}^{\infty} a_n, with the same NN, and the Cauchy criterion for series gives its convergence.

  3. −T≤−Tn≤Sn≤Tn≤T→−T≤S≤T-T \leq -T_n \leq S_n \leq T_n \leq T \rightarrow \resaltar{-T \leq S \leq T}

    Let SS be the sum of the series, TT that of the series of absolute values and TnT_n the partial sums of the latter. Since Tn+1−Tn=∣an+1∣≥0T_{n + 1} - T_n = \abs{a_{n + 1}} \geq 0, the sequence (Tn)(T_n) is increasing, and by the monotone convergence theorem, from the lesson on monotonic sequences and the number e, its limit TT is the supremum of its terms. The extended triangle inequality gives ∣Sn∣≤Tn≤T\abs{S_n} \leq T_n \leq T, and the preservation of inequalities, from the theorem stating that the limit theorems hold for sequences, from the lesson on sequences and their limit, applied to Sn≤TS_n \leq T and to −T≤Sn-T \leq S_n, gives the bound.

This argument already appeared in the course of proving the corollary on the tests for terms of arbitrary sign, from the lesson on the ratio and root tests; here it is stated in general. The series of absolute values has nonnegative terms; by the theorem on series with nonnegative terms, from the lesson on series with positive terms, it converges or diverges to ∞\infty. Therefore, a convergent series converges either absolutely or conditionally, and not in both ways; and a series with nonnegative terms never converges conditionally, because then ∣an∣=an\abs{a_n} = a_n. Conditional convergence requires the positive and the negative terms to cancel each other out, and it is fragile: the lesson on absolute convergence, rearrangements and products of series will prove that, in a conditionally convergent series, the series of the positive terms and that of the negative ones diverge, and that changing the order of the terms can change the sum, which does not happen with an absolutely convergent series. The converse of the theorem is false, as the corollary on the alternating harmonic series, further on, shows.