Part I · Chapter 2

The zeta function

The is the sum ζ(s)=1+1/2s+1/3s+1/4s+⋯\zeta(s) = 1 + 1/2^{s} + 1/3^{s} + 1/4^{s} + \cdots This chapter treats it for real numbers ss only: when the sum has a value, and why it equals a product over the .

The sum

Fix a real number ss and add up 1/ns1/n^{s} over the whole numbers n=1,2,3,…n = 1, 2, 3, \dots For s=2s = 2 the terms are 1,1/4,1/9,1/16,…1, 1/4, 1/9, 1/16, \dots, and the sum is 1.6449… ( showed it is exactly π2/6\pi^{2}/6). The larger ss is, the faster the terms shrink.

For s=1s = 1 the terms are 1,1/2,1/3,1/4,…1, 1/2, 1/3, 1/4, \dots They shrink, but too slowly: the sum of the first NN terms is about ln⁡N+0.577\ln N + 0.577, which grows without bound. This is the harmonic series, and it has no finite sum.

The dividing line is s=1s = 1. To see why, compare each term with an area. The term 1/ns1/n^{s} is the area of a rectangle of width 1 and height 1/ns1/n^{s}. Place that rectangle on the interval from n−1n - 1 to nn. The curve y=1/xsy = 1/x^{s} is decreasing, so on that interval it lies above the rectangle, and the area under the curve is larger:

1ns  ≤  ∫n−1ndxxs\frac{1}{n^{s}} \;\le\; \int_{n-1}^{n} \frac{dx}{x^{s}}

Add this up over n=N+1,N+2,…n = N + 1, N + 2, \dots The intervals join up into everything to the right of NN:

∑n>N1ns  ≤  ∫N∞dxxs  =  N1−ss−1when s>1\sum_{n>N} \frac{1}{n^{s}} \;\le\; \int_{N}^{\infty} \frac{dx}{x^{s}} \;=\; \frac{N^{1-s}}{s-1} \qquad \text{when } s>1

So for s>1s \gt 1 the terms after the NN-th add up to at most N1−s/(s−1)N^{1 - s}/(s - 1), which shrinks to 0 as NN grows. The sum converges.

Placing each rectangle one step to the right, on the interval from nn to n+1n + 1, puts it above the curve and gives the opposite inequality: the terms from the NN-th on add up to at least the same . For s≤1s \le 1 that integral is infinite, so the sum is too. The two bounds differ by less than a single term, so the integral is also a good estimate of what the remaining terms add up to.

So ζ(s)\zeta(s) is defined, for now, only for s>1s \gt 1.

Lab 2.1 adding up the first N terms of 1/ns
the curve y = 1/xsrectangles of height 1/ns for the terms after the N-th
sum of the first n termsdashed line: ζ(s), when s > 1
Top: the next ten terms after the NN-th, each drawn as a rectangle of width 1. They fit under the curve, so their total area is less than the area under the curve. Bottom: the running total. Its horizontal axis is the number of terms, on a . For s>1s \gt 1 the curve levels off at ζ(s)\zeta(s). For s≤1s \le 1 it keeps climbing: at s=1s = 1 it is a straight line on this scale, because the sum grows like ln⁡n\ln n.

Near s=1s = 1

As ss decreases toward 1 the sum grows without bound. The integral above with N=1N = 1 says how: ζ(s)\zeta(s) is close to 1/(s−1)1/(s - 1). More precisely, ζ(s)−1/(s−1)\zeta(s) - 1/(s - 1) tends to 0.5772…, a number called .

Lab 2.2 ζ(s) for real s > 1
ζ(s)1/(s − 1) + 0.5772
Hover to read values. Near s=1s = 1 the two curves coincide. For large ss the sum is barely more than its first term, so ζ(s)\zeta(s) approaches 1.

The product over primes

Recall the geometric series: for ∣x∣<1|x| \lt 1, 1+x+x2+x3+⋯=1/(1−x)1 + x + x^{2} + x^{3} + \cdots = 1/(1 - x). Put x=1/2sx = 1/2^{s}:

11−2−s  =  1+12s+14s+18s+⋯\frac{1}{1-2^{-s}} \;=\; 1+\frac{1}{2^{s}}+\frac{1}{4^{s}}+\frac{1}{8^{s}}+\cdots This is the part of the zeta sum that runs over the powers of 2.

Do the same for 3 and multiply the two series together:

11−2−s⋅11−3−s  =  (1+12s+14s+⋯ )(1+13s+19s+⋯ )\frac{1}{1-2^{-s}}\cdot\frac{1}{1-3^{-s}} \;=\; \left(1+\frac{1}{2^{s}}+\frac{1}{4^{s}}+\cdots\right)\left(1+\frac{1}{3^{s}}+\frac{1}{9^{s}}+\cdots\right) =  1+12s+13s+14s+16s+18s+19s+112s+⋯=\; 1+\frac{1}{2^{s}}+\frac{1}{3^{s}}+\frac{1}{4^{s}}+\frac{1}{6^{s}}+\frac{1}{8^{s}}+\frac{1}{9^{s}}+\frac{1}{12^{s}}+\cdots

Each product of a term from the first bracket and a term from the second is 1/ns1/n^{s} for n=2a3bn = 2^{a}3^{b}, and each such nn occurs once. The numbers 5, 7, 10, 11, … are missing because they have a prime factor other than 2 and 3. Include a factor for every prime, and every whole number appears exactly once, because every whole number is a product of primes in exactly one way. This is Euler's product formula:

ζ(s)  =  ∏p prime11−p−sfor s>1\zeta(s) \;=\; \prod_{p\ \text{prime}} \frac{1}{1-p^{-s}} \qquad \text{for } s>1

Lab 2.3 the product over the primes up to P
product over the primes up to xdashed line: ζ(s)
Each prime multiplies the product by a factor slightly larger than 1, and the steps rise toward ζ(s)\zeta(s). The last readout counts the numbers from 1 to 100 whose prime factors are all at most P: those are the terms of the sum that the truncated product already contains.

Three consequences

−ζ′(s)ζ(s)  =  ∑nΛ(n)ns-\frac{\zeta^{\prime}(s)}{\zeta(s)} \;=\; \sum_{n} \frac{\Lambda(n)}{n^{s}} Here Λ(n)=ln⁡p\Lambda(n) = \ln p if nn is a power of the prime pp, and 0 otherwise.

The left side is built from ζ\zeta alone. The right side is a sum over with the same weights that define ψ(x)\psi(x). Exercise 2 derives it.

Things to try

  • In Lab 2.1, set s=1s = 1 and raise N. Each factor of 10 in N adds the same amount, about 2.3=ln⁡102.3 = \ln 10.
  • Set s=1.1s = 1.1. The sum converges, to about 10.58, but after 10,000 terms it has not reached 7. Compare the "still missing" readout with the bound next to it.
  • In Lab 2.3, set s=2s = 2 and P = 2, 3, 5. The product is 4/3, then 3/2, then 25/16. Watch how many of the numbers up to 100 are covered.
  • Lower ss toward 1.1. Many more primes are needed for the product to come close.

Exercises

  1. Multiply out the for s=2s = 2 using only the primes 2 and 3, and compare with ζ(2)≈1.645\zeta(2) \approx 1.645.
    Solution

    11−14⋅11−1/9=(4/3)(9/8)=1.5\dfrac{1}{1 - \tfrac14} \cdot \dfrac{1}{1 - 1/9} = (4/3)(9/8) = 1.5. Multiplied out, this is the sum of 1/n21/n^{2} over the numbers nn whose only prime factors are 2 and 3. The missing 0.145 comes from every nn divisible by a larger prime.

  2. Derive −ζ′(s)ζ(s)=∑nΛ(n)ns\displaystyle -\frac{\zeta^{\prime}(s)}{\zeta(s)} = \sum_{n}\frac{\Lambda(n)}{n^{s}} from the Euler product.
    Solution

    Start from ln⁡ζ(s)=−∑pln⁡ ⁣(1−p−s)\displaystyle \ln\zeta(s) = -\sum_{p}\ln\!\left(1-p^{-s}\right) and differentiate: ζ′(s)ζ(s)  =  −∑pln⁡p⋅p−s1−p−s  =  −∑pln⁡p⋅(p−s+p−2s+p−3s+⋯ )\frac{\zeta^{\prime}(s)}{\zeta(s)} \;=\; -\sum_{p} \ln p\cdot\frac{p^{-s}}{1-p^{-s}} \;=\; -\sum_{p} \ln p\cdot\left(p^{-s}+p^{-2s}+p^{-3s}+\cdots\right) The second step is the geometric series. The term p−ksp^{-ks} is 1/ns1/n^{s} for n=pkn = p^{k}, with coefficient ln⁡p=Λ(n)\ln p = \Lambda(n).

  3. Use the integral estimate to say how many terms are needed to get ζ(2)\zeta(2) to within 0.001, and ζ(1.1)\zeta(1.1) to within 0.001.
    Solution

    The missing tail is about N1−s/(s−1)N^{1 - s}/(s - 1). For s=2s = 2 that is 1/N1/N, so N≈1,000N \approx 1{,}000. For s=1.1s = 1.1 it is 10⋅N−0.110 \cdot N^{-0.1}, which is below 0.001 only when N>1040N \gt 10^{40}. Close to s=1s = 1 the sum converges, but far too slowly to compute by adding terms.

Check your understanding

  1. The terms 1/n1/n shrink to 0, yet the harmonic series has no finite sum. What decides whether ∑1/ns\sum 1/n^{s} converges?

    • Whether the terms tend to 0
    • How fast they shrink: the tail behaves like ∫dx/xs\int dx/x^{s}, which is finite only for s>1s \gt 1
    • Whether ss is a whole number

    Terms tending to 0 is necessary but not enough. At s=1s = 1 the first NN terms add up to about ln⁡N\ln N, which is unbounded.

  2. The product using only the primes 2 and 3 is, multiplied out, a sum of 1/ns1/n^{s} over which nn?

    • All nn up to 3
    • The multiples of 2 or 3
    • The nn with no prime factor other than 2 and 3: 1, 2, 3, 4, 6, 8, 9, 12, …

    Each term of the expanded product is 1/(2a3b)s1/(2^{a}3^{b})^{s}. 10 is a multiple of 2 but is missing, because it also needs the prime 5.

  3. How does the product formula show there are infinitely many primes?

    • With finitely many primes the product would stay finite as s→1s \to 1, but the sum grows without bound there
    • The product has infinitely many factors by definition
    • It does not; that needs a separate argument

    Sum and product are equal for every s>1s \gt 1. A finite product of the factors 1/(1−p−s)1/(1 - p^{-s}) tends to a finite number as s→1s \to 1, while the sum behaves like 1/(s−1)1/(s - 1). This was Euler's proof.