P Series Calculator: How to Tell If 1/nᵖ Converges or Diverges
About the Author
Marko Šinko
Co-Founder & Lead Developer, AI Math Calculator
Every p series calculator applies the same one-line rule: Σ 1/nᵖ converges when p > 1 and diverges when p ≤ 1. That’s the entire test. No limits to compute, no ratios to simplify — you read the exponent and you’re done. So why does this family of series get its own named test, its own exam questions, and its own calculator? Because the boundary hides more than the rule suggests. The harmonic series (p = 1) diverges even though its terms shrink to zero. Σ 1/n² converges to exactly π²/6, a fact that took the best mathematicians of the 1700s ninety years to prove. And nearly every comparison argument you’ll write in Calc 2 leans on a p-series as its benchmark. This page covers the rule, the reason the cutoff sits at 1, and the places students most often get burned.
The Whole Test in One Sentence
A p-series is any series of the form Σ 1/nᵖ where p is a positive constant. The p-series test says: the series converges if and only if p > 1. So Σ 1/n² converges (p = 2), Σ 1/n³ converges (p = 3), and even Σ 1/n^1.001 converges — barely, with a sum around 1000. On the other side, Σ 1/n diverges (p = 1), Σ 1/√n diverges (p = ½), and anything with p ≤ 0 diverges immediately because its terms don’t even approach zero.
Two details trip people up. First, the boundary itself belongs to the divergent side: p = 1 diverges, full stop. There is no p-series that “converges slowly at p = 1.” Second, the starting index is irrelevant to the verdict. Σ 1/n² from n = 1 and from n = 500 both converge — dropping finitely many terms changes the sum, never the behavior. The calculator above lets you move the starting index precisely so you can watch the sum change while the verdict doesn’t.
Why the Boundary Sits Exactly at p = 1
The rule isn’t arbitrary — it falls straight out of the integral test. Compare Σ 1/nᵖ with the integral ∫₁^∞ 1/xᵖ dx. The series and the integral converge or diverge together, and the integral is elementary calculus:
∫₁^∞ x⁻ᵖ dx = [x^(1−p)/(1−p)]₁^∞
p > 1: exponent 1−p is negative → x^(1−p) → 0 → integral = 1/(p−1) ✓ finite
p < 1: exponent 1−p is positive → x^(1−p) → ∞ ✗ infinite
p = 1: ∫₁^∞ dx/x = ln x → ∞ ✗ infinite (logarithmically)
The exponent 1 − p flips sign exactly at p = 1, and that sign flip is the whole story. When you want to verify this mechanically for a specific function, our integral test calculator runs the same comparison for any positive decreasing term, not just powers of n.
The Harmonic Series: Divergence in Slow Motion
Set p = 1 in the calculator and look at the partial sums. After a million terms, 1 + ½ + ⅓ + … has reached only about 14.39. The sums grow like ln N + 0.5772, which means every additional unit of sum costs about 2.7 times as many terms as the last one. Want the sum to pass 20? You’ll need roughly 2.7 × 10⁸ terms. Past 100? About 1.5 × 10⁴³ terms — more terms than there are atoms in a trillion Earths. It diverges, but on a geological timescale.
| Terms N | Harmonic sum (p = 1) | Σ 1/n² sum (p = 2) |
|---|---|---|
| 10 | 2.929 | 1.5498 |
| 100 | 5.187 | 1.6350 |
| 10,000 | 9.788 | 1.6448 |
| 1,000,000 | 14.393 | 1.64493 |
| Limit | ∞ | π²/6 ≈ 1.64493 |
The side-by-side is the lesson. Both series have terms shrinking to zero, both look tame for the first few dozen terms, and they end up in different universes. Terms going to zero is necessary for convergence, never sufficient — the harmonic series is the standard counterexample, and it’s worth memorizing because exam writers love it.
Exact Sums: the Basel Problem and the Zeta Function
The p-series test tells you whether Σ 1/nᵖ converges, not what it converges to. The sums themselves are values of the Riemann zeta function, ζ(p), and they’re famous. Euler’s 1734 solution of the Basel problem showed ζ(2) = π²/6 — a shock at the time, since nobody expected π to appear in a sum of reciprocal squares. Even exponents all have closed forms in powers of π. Odd exponents don’t: ζ(3) ≈ 1.2020569 (Apéry’s constant) was proved irrational only in 1978, and no closed form is known to this day.
| p | Exact sum ζ(p) | Decimal |
|---|---|---|
| 2 | π²/6 | 1.644934 |
| 3 | no closed form (Apéry’s constant) | 1.202057 |
| 4 | π⁴/90 | 1.082323 |
| 6 | π⁶/945 | 1.017343 |
| 1.5 | no closed form | 2.612375 |
Notice how fast the sums approach 1 as p grows: by p = 6 the entire tail past the first term contributes less than 0.018. Big exponents make the first term do almost all the work.
The Real Job of p-Series: Benchmarks for Every Comparison
Textbook exercises rarely hand you a clean 1/nᵖ. They hand you (3n + 2)/(n³ − 5) and ask you to deal with it. The move is always the same: figure out which p-series your messy series behaves like for large n, then compare. Here, the top grows like n and the bottom like n³, so the terms behave like 1/n² — compare against the convergent p-series with p = 2 using the limit comparison test, get a finite positive limit, and convergence follows.
A quick decision guide for picking the benchmark exponent: keep only the highest power of n on top and bottom, subtract, and that difference is your effective p. Terms like (2n + 7)/(n² + n) act like 1/n (diverges). Terms like 5/(n√n) are 5/n^1.5 (converges). If the result lands exactly on p = 1, expect the problem to be testing whether you remember that the harmonic series diverges. And when the series isn’t a rational function at all — factorials, exponentials, alternating signs — a p-series is the wrong benchmark; reach for the ratio test for factorial growth or the alternating series test when signs flip. Our series convergence calculator picks among all of these automatically if you’d rather start from the general tool.
Near Misses: Σ 1/(n ln n) and the Series That Almost Qualify
Here’s the trap that catches strong students. The series Σ 1/(n ln n) has terms smaller than the harmonic series — surely shrinking the terms of a divergent series can’t hurt? It still diverges. The integral test gives ∫ dx/(x ln x) = ln(ln x), which crawls to infinity slower than anything you’ve plotted, but gets there. To reach a sum of just 10, you’d need around 10^9500 terms. Yet add one more logarithm squared — Σ 1/(n (ln n)²) — and the series converges, to about 2.11.
The lesson: p-series behavior isn’t about terms being “small,” it’s about the exact power of n. Between p = 1 and every p > 1 there’s an infinite hierarchy of logarithmic corrections, and the p-series test says nothing about them — you need the integral test the moment a logarithm shows up. Three specific mistakes to avoid on exams: don’t apply the p-series verdict to Σ 1/(n ln n) (it’s not a p-series); don’t claim Σ 1/n converges “because 1/n → 0”; and don’t confuse p-series with geometric series — Σ 1/n² has the variable in the base, Σ 1/2ⁿ has it in the exponent, and the convergence rules (p > 1 versus |r| < 1) are completely different animals.
