The whole analytic strategy for counting primes runs on a single bet: that facts about a Dirichlet series F(s) = \sum_{n\ge 1} a_n n^{-s} as a complex function — where it has poles, where it stays finite, how it behaves as s approaches the edge of its half-plane of convergence — can be cashed back out as arithmetic facts about the partial sums \sum_{n\le x} a_n. A Tauberian theorem is the exact machine that performs that exchange. Feed it "the function is analytic up to the line \Re(s)=1 except for one simple pole," and it hands back "the coefficient sums grow like Ax." It is the gearbox that turns the Prime Number Theorem's analytic input — \zeta has no zeros on \Re(s)=1 — into the arithmetic output \psi(x)\sim x.

The name honours Alfred Tauber, who in 1897 proved the first theorem of this shape. The label "Tauberian" was coined by G. H. Hardy and J. E. Littlewood, who turned Tauber's one modest result into an entire industry. Understanding what these theorems can and cannot do — and why the hard direction is hard — is the heart of this page.

Two directions: Abelian versus Tauberian

Everything here lives in the tension between two opposite implications. Start with the simplest averaging process, Abel summation: given a series \sum a_n, form the power series f(r) = \sum_{n\ge 0} a_n r^{n} and look at \lim_{r\to 1^-} f(r). If that limit exists we call the series Abel summable.

In words: convergence implies summability. Any genuinely convergent series is also Abel summable, to the same value. This direction is easy — it is a one-line consequence of summation by parts — and results of this shape (regularity: "convergence ⇒ this smoother average agrees") are called Abelian theorems.

The reverse implication is what we actually want, and it is false as stated. A series can be Abel summable without converging at all: the classic offender is Grandi's series 1 - 1 + 1 - 1 + \cdots, whose partial sums bounce 1,0,1,0,\dots forever (no limit), yet \sum (-1)^n r^n = \tfrac{1}{1+r} \to \tfrac12 as r\to 1^-. Summable, not convergent. So you cannot run Abel's theorem backwards for free.

A Tauberian theorem is a partial converse that buys back the reverse implication by paying a price — an extra side condition on the coefficients a_n that rules out the pathological cases. Schematically:

Abelian is the free direction; Tauberian is the direction we crave, and the whole art is finding the weakest side condition that makes the converse true.

Tauber's prototype (1897)

Tauber's own theorem is the cleanest illustration of "summable + a side condition ⇒ convergent." The side condition he chose is that the terms decay a little faster than 1/n.

Notice how surgically the side condition kills the counterexample. For Grandi's series a_n = (-1)^n, so n\,a_n = (-1)^n n does not tend to 0 — the hypothesis fails, and Tauber's theorem simply declines to apply. That is exactly the behaviour we want: the theorem is true precisely because it excludes the series that break it.

Hardy and Littlewood spent decades weakening "o(1/n)" toward the far more useful "O(1/n)" (bounded n a_n) and one-sided variants like a_n \ge -C/n. Every such relaxation is a deeper Tauberian theorem, and the deepest of them power the proofs of the Prime Number Theorem.

The workhorse: the Wiener–Ikehara theorem

For number theory the decisive Tauberian theorem is the one Norbert Wiener and his student Shikao Ikehara proved in 1931–32. Its genius is the side condition it uses: not a decay rate, but simple non-negativity of the coefficients — a condition arithmetic hands us for free, because counting functions are built from non-negative pieces.

Read the hypotheses as the two ingredients of every Tauberian statement. The analytic input is "one simple pole at s=1, otherwise good up to the boundary line." The Tauberian side condition is the single inequality a_n \ge 0. Out drops a clean asymptotic for the partial sums, with the leading constant handed to you as the residue of the pole. Nothing about individual a_n is claimed — only their running total, which is exactly the object arithmetic cares about.

Worked example: Wiener–Ikehara ⇒ the Prime Number Theorem

Here is the payoff the whole toolkit was built for. The right Dirichlet series to feed in is the logarithmic derivative of zeta. Differentiating the Euler product gives, for \Re(s)>1,

where \Lambda(n) is the von Mangoldt function: \Lambda(n) = \ln p if n=p^k is a prime power, and 0 otherwise. Crucially, every \Lambda(n) \ge 0 — the non-negativity hypothesis is satisfied automatically. Now line up the two ingredients.

• The pole. Since \zeta(s) has a simple pole at s=1, it looks like \zeta(s)\approx \tfrac{1}{s-1} there, so -\zeta'/\zeta has a simple pole at s=1 with residue exactly A = 1. (A simple pole of \zeta becomes a simple pole of its logarithmic derivative with residue equal to the pole's order, 1.)
• Analytic to the boundary. A simple pole of -\zeta'/\zeta also appears at every zero of \zeta. The one deep analytic fact — the same fact behind every proof of the PNT — is that \zeta(s)\ne 0 on the line \Re(s)=1. That is precisely what guarantees no extra poles sit on the boundary, so -\zeta'/\zeta - \tfrac{1}{s-1} extends to \Re(s)\ge 1.

Both hypotheses hold with A=1, so Wiener–Ikehara fires and delivers

This is the Prime Number Theorem in its natural analytic form. A short partial-summation argument converts \psi(x)\sim x into the familiar \pi(x)\sim x/\ln x. The entire logical distance from "no zeros on the edge line" to "primes thin out like 1/\ln x" is bridged by that one Tauberian step.

The modern short route: Newman's analytic Tauberian theorem

Wiener's original proof leaned on hard Fourier analysis. In 1980 Donald J. Newman found a startlingly short proof of the same class of results using nothing but Cauchy's theorem and a clever contour — the version popularised by Don Zagier in a famous three-page write-up, "The simplest proof of the prime number theorem." It is now the standard graduate route.

The engine of the proof is a single contour integral. To recover the tail of the series one writes a difference like

where the ingenious extra factor \bigl(\tfrac1z + \tfrac{z}{R^{2}}\bigr) — Newman's kernel — vanishes on |z|=R in just the way needed to make the arcs of the contour contribute negligibly. The contour C is the boundary of a disc of radius R centred at the point of interest, truncated on the left by a short vertical segment at \Re(z) = -\delta, so that it can be pushed just past the boundary line into the region where g is now known to be analytic. That truncation is the entire trick: analyticity up to the line lets the contour bulge left of it, Cauchy's theorem collects the pole, and careful bounds on the pieces make everything else disappear as R\to\infty.

The sketch shows the shape: a big circle of radius R, sliced by the vertical line \Re(z)=-\delta, with the contour hugging the boundary of the truncated region. The pole to be captured sits on the imaginary axis; the left bulge is only possible because the Tauberian hypothesis gave us analyticity a hair past the critical line.

The pattern, in one line

Every application in this course fits the same template. Encode your arithmetic in a Dirichlet series; prove it continues to the boundary line with a controlled pole; supply a Tauberian side condition (usually non-negativity, handed to you for free); read off the asymptotic from the residue.

The middle row is a sanity check — feeding \zeta itself back through the machine recovers the trivial \lfloor x\rfloor\sim x. The bottom row is a preview of what higher-order poles buy you: a double pole promotes the main term from Ax to Ax\ln x. The residue calculus, not the arithmetic, is doing the bookkeeping.