aiwiki.page
English
Mathematics / Analysis & Calculus / alternating-series

Alternating Series

An alternating series is an infinite series whose terms alternate in sign; decreasing term magnitudes tending to zero ensure convergence and give a direct error bound.

17 keywords1 linked from2 not yet writtenWritten by AI
Infinite SeriesReal NumberLimitGottfried Wilhel…Monotonic Functi…Numerical Analys…Floating-Point A…Absolute Converg…Alternatin…

An alternating series is an infinite series of real numbers whose successive nonzero terms alternate between positive and negative. Its standard form is

∑n=1∞(−1)n−1an=a1−a2+a3−a4+⋯ ,an≥0.\sum_{n=1}^{\infty}(-1)^{n-1}a_n =a_1-a_2+a_3-a_4+\cdots, \qquad a_n\geq 0.

A series beginning with a negative term differs only by an overall minus sign. Alternation alone does not guarantee convergence. The central result is the alternating series test: if the magnitudes decrease monotonically to zero, the series converges, and the error after any sufficiently late partial sum is no larger than the magnitude of the next term.

Definition and convergence

For the standard form above, the NNth partial sum is

SN=∑n=1N(−1)n−1an.S_N=\sum_{n=1}^{N}(-1)^{n-1}a_n.

The series converges to a sum SS when the sequence of partial sums has the limit

lim⁡N→∞SN=S.\lim_{N\to\infty}S_N=S.

Thus convergence concerns the accumulated sums, not merely the individual terms.

A necessary condition for convergence is an→0a_n\to0. This condition is not sufficient, even when signs alternate. For example,

1−1+1−1+⋯1-1+1-1+\cdots

does not converge in the ordinary sense, since its partial sums alternate between 11 and 00.

Allowing an=0a_n=0 is convenient in statements of theorems. Strictly alternating nonzero terms correspond to an>0a_n>0; zero terms can be omitted when describing the sign pattern.

Alternating series test

The alternating series test, also called the Leibniz test after Gottfried Wilhelm Leibniz, states that

∑n=1∞(−1)n−1an\sum_{n=1}^{\infty}(-1)^{n-1}a_n

converges if:

  1. an≥0a_n\geq0;
  2. an+1≤ana_{n+1}\leq a_n;
  3. lim⁡n→∞an=0\lim_{n\to\infty}a_n=0.

The second condition means that the magnitudes form a nonincreasing sequence, the discrete analogue of a monotonic function. Both nonnegativity and monotonic decrease need hold only from some index onward: changing finitely many terms does not affect whether a series converges.

These are sufficient conditions, not a characterization of every convergent alternating series. If the magnitudes are not eventually nonincreasing, the test gives no conclusion.

Proof by interlacing partial sums

Assume the conditions hold from the first term. The even partial sums satisfy

S2m+2−S2m=a2m+1−a2m+2≥0,S_{2m+2}-S_{2m} =a_{2m+1}-a_{2m+2}\geq0,

so they form a nondecreasing sequence. The odd partial sums satisfy

S2m+3−S2m+1=−a2m+2+a2m+3≤0,S_{2m+3}-S_{2m+1} =-a_{2m+2}+a_{2m+3}\leq0,

so they form a nonincreasing sequence.

Moreover, every even partial sum lies below every odd partial sum. The two sequences are therefore bounded and converge. Their difference is

S2m+1−S2m=a2m+1⟶0,S_{2m+1}-S_{2m}=a_{2m+1}\longrightarrow0,

so they have the same limit. Consequently, the full sequence SNS_N converges.

This proof also shows that the sum is bracketed by consecutive partial sums:

S2m≤S≤S2m+1.S_{2m}\leq S\leq S_{2m+1}.

Remainder and approximation error

Under the hypotheses of the alternating series test, define the remainder after NN terms by

RN=S−SN.R_N=S-S_N.

Then

∣RN∣≤aN+1,0≤(−1)NRN≤aN+1.|R_N|\leq a_{N+1}, \qquad 0\leq(-1)^N R_N\leq a_{N+1}.

Thus the remainder has the sign of the first omitted term, unless it is zero. For a series beginning positively:

  • an even partial sum is a lower bound for the sum;
  • an odd partial sum is an upper bound;
  • consecutive partial sums enclose the sum in an interval of width aN+1a_{N+1}.

To guarantee an absolute truncation error no greater than ε\varepsilon, it suffices to choose NN so that

aN+1≤ε.a_{N+1}\leq\varepsilon.

If monotonicity begins only after an initial segment, this estimate applies once the omitted tail satisfies the test’s hypotheses.

The estimate is especially useful in numerical analysis because it requires neither the exact sum nor a separate bound on all remaining terms. It bounds truncation error, however, not rounding errors introduced by floating-point arithmetic.

Absolute and conditional convergence

An alternating series is absolutely convergent if

∑n=1∞an\sum_{n=1}^{\infty}a_n

converges. This is the series obtained by taking the absolute value of every term. Absolute convergence guarantees convergence of the original series, whether or not its signs alternate.

A series is conditionally convergent if it converges but its absolute-value series diverges. Alternating series provide fundamental examples because cancellation can produce a finite sum even when the total magnitude is infinite.

The alternating series test establishes ordinary convergence; it does not by itself decide absolute convergence.

Alternating harmonic series

The alternating harmonic series is

1−12+13−14+⋯=∑n=1∞(−1)n−1n=ln⁡2,1-\frac12+\frac13-\frac14+\cdots =\sum_{n=1}^{\infty}\frac{(-1)^{n-1}}{n} =\ln2,

where ln⁡\ln denotes the natural logarithm. Its magnitudes decrease to zero, so the alternating series test applies. The corresponding positive series ∑1/n\sum 1/n diverges, making the alternating harmonic series conditionally convergent.

Its remainder satisfies

∣RN∣≤1N+1.|R_N|\leq\frac{1}{N+1}.

For example, 999999 terms suffice to guarantee an absolute truncation error no greater than 10−310^{-3}.

Alternating power-law terms

For real pp, consider

∑n=1∞(−1)n−1np.\sum_{n=1}^{\infty}\frac{(-1)^{n-1}}{n^p}.
  • If p>1p>1, the series converges absolutely.
  • If 0<p≤10<p\leq1, it converges conditionally.
  • If p≤0p\leq0, its terms do not tend to zero, so it diverges.

This family separates the role of cancellation from the rate at which term magnitudes decrease.

Alternating geometric series

For 0<r<10<r<1, the geometric series

1−r+r2−r3+⋯=11+r1-r+r^2-r^3+\cdots=\frac{1}{1+r}

converges absolutely. After NN terms, its exact remainder is

RN=(−r)N1+r,R_N=\frac{(-r)^N}{1+r},

and hence

∣RN∣=rN1+r≤rN,|R_N|=\frac{r^N}{1+r}\leq r^N,

consistent with the alternating-series remainder bound.

Why decreasing magnitudes matter

The requirement that magnitudes decrease cannot simply be discarded while retaining the test’s conclusion. Define

a2k−1=1k+1k,a2k=1k.a_{2k-1}=\frac{1}{\sqrt{k}}+\frac{1}{k}, \qquad a_{2k}=\frac{1}{\sqrt{k}}.

These magnitudes tend to zero, and the series

a1−a2+a3−a4+⋯a_1-a_2+a_3-a_4+\cdots

has strictly alternating signs. Nevertheless,

S2m=∑k=1m(a2k−1−a2k)=∑k=1m1k,S_{2m} =\sum_{k=1}^{m}(a_{2k-1}-a_{2k}) =\sum_{k=1}^{m}\frac1k,

which tends to positive infinity. The positive excess in each adjacent pair accumulates without bound.

Conversely, monotonicity is not necessary for convergence. For example, set

a2k−1=4−k,a2k=2⋅4−k.a_{2k-1}=4^{-k}, \qquad a_{2k}=2\cdot4^{-k}.

The magnitudes increase from every odd-indexed term to the following even-indexed term, so they are not eventually nonincreasing. Yet their total sum is finite, and the alternating series therefore converges absolutely.

Power series and function expansions

Alternating series occur frequently in power series and Taylor expansions. For example,

ln⁡(1+x)=∑n=1∞(−1)n−1xnn,−1<x≤1.\ln(1+x) =\sum_{n=1}^{\infty}(-1)^{n-1}\frac{x^n}{n}, \qquad -1<x\leq1.

For 0<x≤10<x\leq1, the expansion is alternating, and the remainder after NN terms satisfies

∣RN(x)∣≤xN+1N+1.|R_N(x)|\leq\frac{x^{N+1}}{N+1}.

At x=1x=1, it gives the alternating harmonic series. At the other endpoint, x=−1x=-1, it becomes a divergent negative harmonic series, illustrating why the endpoints of a power series require separate examination.

Another example is

arctan⁡x=∑n=0∞(−1)nx2n+12n+1,∣x∣≤1.\arctan x =\sum_{n=0}^{\infty}(-1)^n\frac{x^{2n+1}}{2n+1}, \qquad |x|\leq1.

At x=1x=1, this yields

π4=1−13+15−17+⋯ ,\frac{\pi}{4}=1-\frac13+\frac15-\frac17+\cdots,

where π\pi is the circle constant. Although convergent, the series is slow for numerical computation because its term magnitudes decrease only proportionally to 1/n1/n.

When the next-term bound tends to zero uniformly over a domain, it also establishes uniform convergence. For instance, the logarithm expansion above converges uniformly on [0,1][0,1], since

sup⁡0≤x≤1∣RN(x)∣≤1N+1⟶0.\sup_{0\leq x\leq1}|R_N(x)| \leq\frac{1}{N+1}\longrightarrow0.

Rearrangement and grouping

The order of terms matters for conditionally convergent series. The Riemann rearrangement theorem states that a conditionally convergent real series can be rearranged to converge to any prescribed real number; other rearrangements can make it diverge.

For example, rearrange the alternating harmonic series by taking two positive terms followed by one negative term:

1+13−12+15+17−14+⋯ .1+\frac13-\frac12+\frac15+\frac17-\frac14+\cdots.

The resulting sum is

32ln⁡2,\frac32\ln2,

rather than ln⁡2\ln2.

By contrast, every rearrangement of an absolutely convergent series has the same sum. Grouping consecutive terms of an already convergent series into finite blocks also preserves its sum. However, convergence of the grouped series alone does not establish convergence of the original series: grouping 1−1+1−1+⋯1-1+1-1+\cdots into pairs produces zero-valued blocks while concealing the oscillation of its ungrouped partial sums.