Normality of π

$\pi$ begins $3.1415926535\ldots$. Among the first ten digits after the decimal point, $5$ appears three times and $0$ does not appear at all. That small sample is uneven.

Call a consecutive pattern a digit block. A normal number has a particular kind of long-run balance: each digit's share approaches one tenth, each specified pair's share approaches one hundredth, and every longer fixed block has its corresponding share. Does the decimal expansion of $\pi$ have this property?[1] [2]

Does $\pi$ have the long-run digit balance called normality? We do not know. For a decimal normal number, each digit's share approaches one tenth, and each chosen two-digit pattern's share approaches one hundredth, as we examine longer opening stretches of its digits.

Every fixed length must follow its corresponding rule; single-digit balance is not enough.

Take the pattern $00$. At each digit after the decimal point, inspect that digit and the next one. Count a match when both are zero, allowing the pairs to overlap: in the hypothetical string $000$, the pattern starts at the first and second positions. Divide the number of matches by the number of starting positions checked. Normality would mean this share approaches one hundredth, its limiting frequency. It need not equal one hundredth in any finite stretch.

Here is compact bookkeeping for that same test. Write $d_j$ for digit number $j$ after the decimal point, starting with $d_1=1$ in $\pi$. Call the chosen digit block $w$, and its length $k$; for $00$, $k=2$.

Let $N$ be the number of starting positions checked, and $C_w(N)$ the count of matches. The starts run from $1$ to $N$; checking a length-$k$ pattern uses the first $N+k-1$ fractional digits.

The fraction $C_w(N)/N$ is the counted share. The formula says it must approach $10^{-k}$ as $N$ grows without bound, as the arrow indicates. This value means one divided by ten to the power $k$: one tenth for a digit, one hundredth for a pair. Fix each pattern first; the condition must hold for every length and every pattern of that length. Finite counts cannot prove these infinite limits, which remain unproved for $\pi$:

$$\lim_{N\to\infty}\frac{C_w(N)}{N}=10^{-k}$$[1] [2] [3]

We cannot yet prove that each decimal digit's share approaches one tenth in $\pi$, let alone prove the right share for every pair, triple and longer pattern.

Computations examine enormous finite stretches and find evidence consistent with balance. But checking more digits alone cannot settle what happens over the entire infinite expansion. A proof must explain the long-run frequencies.[1]

No proof establishes these limits for $\pi$. The case $k=1$ is already unresolved. Neither a large finite digit census nor a successful numerical calculation using those digits controls their infinite continuation.

The 2012 paper's finite-string tests and statistical model, and the 2026 Monte Carlo experiment, address empirical behavior under their respective assumptions. They do not establish the displayed infinite limits. This page asks for base-ten normality; normality in every integer base is a stronger target.[1] [2]

$\pi$ comes from circles, not from tossing dice. Yet its digits seem to have the balance we expect from repeated random choices. Understanding why a number defined by geometry could have that balance is the appeal of this problem.

A proof would connect $\pi$'s exact mathematical definition with statistical laws for its digits.[3]

The interest is in connecting explicit formulas with statistical behavior. Almost every real number is a normal number, but that theorem does not establish normality for particular constants such as $\pi$. Proving it here would explain how a number fixed by geometry exhibits the digit frequencies associated with independent, equally likely choices.

There is also a connection to dynamics, the study of repeated rules: take the fractional part of $10^n\pi$, for $n=0,1,2,\ldots$. Decimal normality is equivalent to those points being uniformly distributed in $[0,1)$. This connects the problem to methods for studying repeated transformations. Frequency balance still leaves $\pi$'s digits computable; normality does not mean unpredictability.[3]