Fix an elliptic curve over the rationals — say E: y^2 = x^3 - x + 1 — and start counting. For each prime p you reduce the curve modulo p and count how many points it has over the finite field \mathbb{F}_p. The "error" against the naive guess of p+1 points is a single integer,

the trace of Frobenius. As p runs over the primes these a_p jump around, seemingly at random — sometimes positive, sometimes negative, small for some primes and large for others. The Sato–Tate distribution is the astonishing answer to a simple question: if you normalise them properly and make a histogram, what shape does it settle into? The answer is a perfect, symmetric semicircle-flavoured bell — the same curve for every non-CM curve on Earth — and proving it took some of the deepest machinery in modern number theory. This page builds on analytic number theory and the theory of elliptic-curve L-functions.

From the Hasse bound to an angle

The first miracle is that a_p is never large. Helmut Hasse proved in the 1930s that the point count can never stray far from p+1:

This is exactly the statement that the "Riemann Hypothesis for elliptic curves over finite fields" holds — the two Frobenius eigenvalues have absolute value \sqrt{p}. Because a_p is trapped in [-2\sqrt{p}, 2\sqrt{p}], we can divide it out and land in [-1, 1], which is exactly the range of a cosine. So we define an angle \theta_p \in [0, \pi] by

The Frobenius eigenvalues are then \sqrt{p}\,e^{\pm i\theta_p} — a conjugate pair on the circle of radius \sqrt{p}. The number \theta_p is called the Frobenius angle (or Hecke angle, for a modular form). A large positive a_p means \theta_p near 0; a large negative a_p means \theta_p near \pi; a_p \approx 0 means \theta_p \approx \pi/2.

Worked example. Take p = 11 and suppose a_{11} = 4. Then 2\sqrt{11} \approx 6.633, so \cos\theta_{11} = 4/6.633 \approx 0.603 and \theta_{11} = \arccos(0.603) \approx 0.921 radians (about 52.8^\circ) — a fairly small angle, reflecting that a_{11} is comfortably positive.

The conjecture that became a theorem

Mikio Sato (from numerical experiments) and John Tate (from a conceptual, group-theoretic argument) independently arrived at the same guess in the 1960s: those angles are not just bouncing around arbitrarily — they follow a precise, universal law.

Long a conjecture, this is now a theorem. For elliptic curves over \mathbb{Q} (and totally real fields) it was proved around 2008–2011 by Laurent Clozel, Michael Harris, Nicholas Shepherd-Barron and Richard Taylor, and extended to holomorphic modular forms (Hecke eigenforms) shortly after by Barnet-Lamb, Geraghty, Harris and Taylor. The density \tfrac{2}{\pi}\sin^2\theta is zero at the endpoints, rises to a maximum of \tfrac{2}{\pi} at \theta = \pi/2, and is symmetric about the middle — a smooth hump. In the variable x = a_p/\sqrt{p} = 2\cos\theta \in [-2,2] the very same law becomes the semicircle density \tfrac{1}{2\pi}\sqrt{4 - x^2}, which is why people call it "the Sato–Tate semicircle."

Seeing the density

Below is the Sato–Tate density \tfrac{2}{\pi}\sin^2\theta on [0,\pi] (the smooth curve), with a "staircase" overlay that mimics what a histogram of the first few thousand Frobenius angles of a real non-CM curve looks like — the bars hug the curve ever more closely as you include more primes. Notice how the angles avoid the extremes near 0 and \pi and pile up in the middle: it is rare for |a_p| to come close to its Hasse ceiling 2\sqrt{p}, and common for a_p to be modest.

The total area under the curve is 1 — as it must be, since it is a probability density. You can check the normalisation directly: \frac{2}{\pi}\int_0^\pi \sin^2\theta\,d\theta = \frac{2}{\pi}\cdot\frac{\pi}{2} = 1, using \int_0^\pi \sin^2\theta\,d\theta = \pi/2.

Worked example: how often is a_p large and positive?

Let's use the law to make a genuine prediction. "a_p large and positive" means \theta_p is small — close to 0. Take the window \theta_p \in [0, \pi/3] (i.e. \cos\theta_p \ge \tfrac12, so a_p \ge \sqrt{p}). The predicted proportion of such primes is

Use the identity \sin^2\theta = \tfrac12(1 - \cos 2\theta), whose antiderivative is \tfrac{\theta}{2} - \tfrac{\sin 2\theta}{4}:

Numerically \frac{\pi}{6} \approx 0.5236 and \frac{\sqrt3}{8} \approx 0.2165, so P \approx \frac{2}{\pi}(0.3071) \approx 0.1955. About 19.5% of primes should have a_p \ge \sqrt{p}. By symmetry the same 19.5\% have a_p \le -\sqrt{p} (angle in [2\pi/3, \pi]), leaving roughly 61\% of primes with |a_p| < \sqrt{p} — a firm, checkable prediction you can test against a table of any non-CM curve, and it holds beautifully.

Where the measure comes from: SU(2)

Why this curve and not some other? Tate's insight was that the law is pure representation theory. Attach to each prime the conjugacy class of the (unitarised) Frobenius inside the compact group SU(2) of 2\times 2 special unitary matrices. A conjugacy class in SU(2) is determined by a single angle \theta \in [0,\pi] (the eigenvalues are e^{\pm i\theta}), and the pushforward of Haar measure — the natural, rotation-invariant probability measure on the group — onto that angle is exactly

This is the Weyl integration formula for SU(2): the factor \sin^2\theta is the Weyl measure. So the Sato–Tate density is not an analytic accident — it is the shadow of a compact Lie group. That group is called the Sato–Tate group of the curve; for a non-CM elliptic curve it is the full SU(2), and the theorem says the Frobenius classes are equidistributed in it.

The proof strategy: a PNT for every moment

Equidistribution problems have a standard translation into analysis, via Weyl's criterion: the angles equidistribute with respect to \mu_{ST} if and only if, for every "test harmonic," the average of that harmonic tends to its mean. The natural harmonics for SU(2) are the characters of its irreducible representations \operatorname{Sym}^n, which are the Chebyshev-like functions

(these have mean zero against \mu_{ST} for n \ge 1). Feeding \operatorname{Sym}^n of the Frobenius into an Euler product builds the symmetric-power L-functions L(s, \operatorname{Sym}^n E). The whole theorem then reduces to a single analytic input, one for each n:

Getting that holomorphy and non-vanishing for all n at once is the hard part, and it is where the modern machine enters. One proves the L(s, \operatorname{Sym}^n E) are well-behaved by showing each \operatorname{Sym}^n is (potentially) automorphic — that it matches the L-function of an automorphic representation, whose analytic continuation and non-vanishing are known. This is the theory of Hecke eigenforms and their L-functions pushed to its limit.

The CM case is different

The Sato–Tate law as stated is only for curves without complex multiplication. A curve with CM (its endomorphism ring is bigger than \mathbb{Z} — an order in an imaginary quadratic field) obeys a completely different law. For roughly half the primes (the "supersingular" or inert primes) one has a_p = 0 exactly, so \theta_p = \pi/2 sits as an atom of mass \tfrac12. Over the remaining (split) primes the angle is uniform on [0,\pi] — density \tfrac{1}{2\pi} for the continuous part — rather than the \sin^2 hump.

Group-theoretically, the Sato–Tate group of a CM curve is not SU(2) but its normaliser of a maximal torus N(U(1)) — a smaller group, whose Haar measure is uniform on angles. The shape of the histogram literally detects whether or not the curve has complex multiplication.