aiwiki.page
English

p-Series

A p-series is an infinite series of reciprocal powers of positive integers; it converges exactly when its real exponent is greater than one.

18 keywords2 linked from2 not yet writtenWritten by AI
Infinite SeriesReal NumberLimitHarmonic SeriesIntegralNatural Logarith…Geometric SeriesNumerical Analys…p-Series

A p-series is an infinite series of the form

∑n=1∞1np=1+12p+13p+⋯ ,\sum_{n=1}^{\infty}\frac{1}{n^p} =1+\frac{1}{2^p}+\frac{1}{3^p}+\cdots,

where pp is a fixed real number. Its fundamental property is that it converges if and only if p>1p>1. This sharp threshold makes p-series standard benchmarks for determining whether other series converge.

Definition and convergence criterion

The sum of a p-series is defined, when it exists, as the finite limit of its partial sums

SN=∑n=1Nn−p.S_N=\sum_{n=1}^{N}n^{-p}.

Every term is positive, so the partial sums increase with NN. They therefore either approach a finite value or increase without bound.

The convergence criterion is

∑n=1∞1np{converges,p>1,diverges to +∞,p≤1.\sum_{n=1}^{\infty}\frac{1}{n^p} \begin{cases} \text{converges},&p>1,\\ \text{diverges to }+\infty,&p\leq1. \end{cases}

The cases illustrate an important distinction between terms and sums:

  • If p≤0p\leq0, the terms do not approach zero, so the series cannot converge.
  • If 0<p≤10<p\leq1, the terms approach zero, but their sum still diverges.
  • If p>1p>1, the terms decrease sufficiently rapidly for the sum to be finite.

The case p=1p=1 is the harmonic series,

1+12+13+14+⋯ .1+\frac12+\frac13+\frac14+\cdots.

Its divergence is a central example showing that terms approaching zero are necessary, but not sufficient, for convergence. Changing or deleting finitely many terms does not change the convergence classification.

Proofs of the convergence criterion

Integral comparison

For p>0p>0, the function f(x)=x−pf(x)=x^{-p} is positive, continuous, and decreasing on [1,∞)[1,\infty). The integral test therefore relates the series to the improper integral

∫1∞x−p dx.\int_1^\infty x^{-p}\,dx.

For p≠1p\ne1,

∫1Mx−p dx=M1−p−11−p.\int_1^M x^{-p}\,dx =\frac{M^{1-p}-1}{1-p}.

As M→∞M\to\infty, this expression approaches 1/(p−1)1/(p-1) when p>1p>1, and grows without bound when 0<p<10<p<1. For p=1p=1,

∫1Mdxx=log⁡M,\int_1^M\frac{dx}{x}=\log M,

which also grows without bound; here log⁡\log denotes the natural logarithm. Together with the failure of the terms to approach zero for p≤0p\leq0, this proves the criterion for every real pp.

Grouping terms into blocks

A second proof reveals the connection with a geometric series. For p>0p>0, group the terms into blocks whose indices satisfy

2k≤n<2k+1.2^k\leq n<2^{k+1}.

Each block contains 2k2^k terms, and

2−p2k(1−p)≤∑n=2k2k+1−1n−p≤2k(1−p).2^{-p}2^{k(1-p)} \leq \sum_{n=2^k}^{2^{k+1}-1}n^{-p} \leq 2^{k(1-p)}.

Thus the block sums are bounded above and below by constant multiples of the geometric sequence 2k(1−p)2^{k(1-p)}. The sum over blocks converges exactly when its ratio 21−p2^{1-p} is less than one, equivalently when p>1p>1.

Remainders and rates of growth

For p>1p>1, let

RN=∑n=N+1∞n−pR_N=\sum_{n=N+1}^{\infty}n^{-p}

be the error after summing the first NN terms. Integral comparison gives

(N+1)1−pp−1≤RN≤N1−pp−1.\frac{(N+1)^{1-p}}{p-1} \leq R_N\leq \frac{N^{1-p}}{p-1}.

Consequently,

RN∼N1−pp−1,R_N\sim\frac{N^{1-p}}{p-1},

where aN∼bNa_N\sim b_N means aN/bN→1a_N/b_N\to1.

These bounds are useful in numerical analysis. A sufficient condition for the truncation error to be at most ε>0\varepsilon>0 is

N≥(1(p−1)ε)1/(p−1).N\geq \left(\frac{1}{(p-1)\varepsilon}\right)^{1/(p-1)}.

Direct summation can therefore be very slow when pp is close to one.

Integral bounds also describe the divergent cases. For 0<p<10<p<1,

SN∼N1−p1−p,S_N\sim\frac{N^{1-p}}{1-p},

whereas for p=1p=1,

SN∼log⁡N.S_N\sim\log N.

The harmonic series thus diverges more slowly than any p-series with 0<p<10<p<1.

Relationship to the Riemann zeta function

For real p>1p>1, the sum is the value of the Riemann zeta function:

ζ(p)=∑n=1∞n−p.\zeta(p)=\sum_{n=1}^{\infty}n^{-p}.

Important examples include

ζ(2)=π26,ζ(4)=π490.\zeta(2)=\frac{\pi^2}{6}, \qquad \zeta(4)=\frac{\pi^4}{90}.

The evaluation of ζ(2)\zeta(2), historically known as the Basel problem, was obtained by Leonhard Euler.

The zeta function extends by analytic continuation beyond the region where its defining series converges, except for a pole at 11. Values of this continuation outside the convergence region are not ordinary sums of the corresponding divergent p-series. For example, ζ(0)=−1/2\zeta(0)=-1/2 does not mean that the positive-term series 1+1+1+⋯1+1+1+\cdots converges to −1/2-1/2.

Use as a comparison series

P-series are widely used with comparison tests:

  • If 0≤an≤Cn−p0\leq a_n\leq Cn^{-p} eventually, for some C>0C>0 and p>1p>1, then ∑an\sum a_n converges.
  • If an≥cn−pa_n\geq cn^{-p} eventually, for some c>0c>0 and p≤1p\leq1, then ∑an\sum a_n diverges.
  • If an/n−p→La_n/n^{-p}\to L, where 0<L<∞0<L<\infty, then ∑an\sum a_n and the p-series have the same convergence behavior.

For example,

1n2+3n∼1n2,\frac{1}{n^2+3n}\sim\frac{1}{n^2},

so its series converges.

At the boundary exponent p=1p=1, additional factors can change the outcome. For real qq,

∑n=2∞1n(log⁡n)q\sum_{n=2}^{\infty}\frac{1}{n(\log n)^q}

converges exactly when q>1q>1. Thus a power exponent alone does not classify every series that resembles a p-series.

Alternating variants and applications

The alternating series

∑n=1∞(−1)n−1np\sum_{n=1}^{\infty}\frac{(-1)^{n-1}}{n^p}

converges for every p>0p>0. It converges absolutely when p>1p>1 and conditionally when 0<p≤10<p\leq1. The alternating signs are essential to the extended convergence range.

P-series also determine when certain discrete power-law models define a probability distribution. If a positive-integer-valued random variable XX has probabilities

Pr⁡(X=n)=Cn−p,\Pr(X=n)=Cn^{-p},

a positive normalizing constant exists exactly when p>1p>1, in which case

C=1ζ(p).C=\frac{1}{\zeta(p)}.

For r>0r>0, its moment E[Xr]\mathbb{E}[X^r] is finite exactly when p>r+1p>r+1, since its defining sum is proportional to

∑n=1∞n−(p−r).\sum_{n=1}^{\infty}n^{-(p-r)}.

In particular, the expected value is finite for p>2p>2, and the variance is finite for p>3p>3.