By 1850 everyone believed the primes thin out like x/\ln x — Gauss and Legendre had guessed it from tables half a century earlier — but nobody could prove a single honest inequality about \pi(x). Pafnuty Chebyshev broke the deadlock. Using nothing more exotic than the prime factorisation of a factorial and a clever look at a middle binomial coefficient, he showed in 1852 that \pi(x) is pinned to the right order of magnitude: there are real constants close to 1 so that

This is a genuine theorem, forty years ahead of the Prime Number Theorem, and it needs no complex analysis at all. It doesn't say the ratio converges — that was still out of reach — but it traps it in a narrow band and, tantalisingly, shows that if a limit exists it can only be 1. This page walks the whole elementary argument.

The right object to count: ψ(x), not π(x)

Counting primes directly is awkward because a logarithm sits invisibly inside every prime. The trick that makes everything additive is to weight each prime power by \ln p. That is the von Mangoldt function \Lambda(n), and its running total is Chebyshev's psi function:

Read the middle sum aloud: add \ln p once for every prime power p^k (with k\ge 1) that does not exceed x. There is a close cousin, the theta function \vartheta(x) = \sum_{p \le x}\ln p, which keeps only the first power k=1. The two differ by a negligible amount — the higher prime powers contribute only O(\sqrt{x}\,\ln x) — so bounding one bounds the other.

The bridge from \vartheta to \pi is partial summation: since \vartheta(x)=\sum_{p\le x}\ln p and each \ln p \le \ln x, we get \vartheta(x) \le \pi(x)\ln x immediately, and a matching bound in the other direction. So we may forget primes for a while and just bound \psi(x) — a purely additive quantity.

The engine: how primes divide a factorial

Chebyshev's whole argument runs on one classical fact about factorials, due to Legendre. Ask: to what power does a prime p divide n!?

The reasoning is a tidy piece of counting. Among 1,2,\dots,n exactly \lfloor n/p\rfloor are multiples of p — each contributes at least one factor of p. Of those, \lfloor n/p^2\rfloor are multiples of p^2 and contribute a second factor, and so on up the tower. Summing the layers gives the formula. For example, v_2(10!) = 5+2+1 = 8, so 2^8 \parallel 10!.

Now take logarithms of n! = \prod_{p \le n} p^{\,v_p(n!)}:

The double sum on the right is almost \psi in disguise — indeed one can show \ln(n!) = \sum_{m\le n}\psi(n/m). That link between a factorial we can estimate and the \psi we want to bound is the entire secret.

The clever choice: the central binomial coefficient

Rather than n! alone, Chebyshev looked at the central binomial coefficient

Why this one? Two facts about its size clamp it from both sides. First, it is one term of the expansion \sum_{j=0}^{2n}\binom{2n}{j} = 2^{2n} = 4^{n}, and it is the largest of the 2n+1 terms, so

Taking logs, \ln\binom{2n}{n} = (2\ln 2)\,n + O(\ln n). So its logarithm is essentially (2\ln 2)n \approx 1.386\,n — a clean, known quantity. Second, by Legendre's formula the exponent of p in it is

Each bracket \lfloor 2y\rfloor - 2\lfloor y\rfloor is either 0 or 1 — never more. And the only powers p^k that can contribute are those with p^k \le 2n, so at most \lfloor \ln(2n)/\ln p\rfloor layers survive. Hence v_p\binom{2n}{n}\,\ln p \le \ln(2n) for each prime, and

Squeezing ψ from both sides

Combining the two facts about \binom{2n}{n} gives both halves of Chebyshev's estimate.

Lower bound. From \binom{2n}{n} \ge 4^{n}/(2n+1) and the fact that \ln\binom{2n}{n} \le \psi(2n) (every prime power dividing it is \le 2n, and each contributes at most its full \ln p weight, which \psi(2n) already counts),

Dividing by 2n, we find \psi(2n)/(2n) \ge \ln 2 - o(1), so \liminf \psi(x)/x \ge \ln 2 \approx 0.693. A little more care (using \vartheta and summing a telescoping bound) sharpens the lower constant to Chebyshev's celebrated

Upper bound. Consider the difference \psi(2n) - \psi(n): it counts (with \ln p weight) exactly the prime powers in the half-open range (n, 2n]. Every prime p with n < p \le 2n divides (2n)! once but does not divide (n!)^2 at all, so it divides \binom{2n}{n}. Therefore

Feeding this back at the dyadic scales n, n/2, n/4,\dots and adding the telescoping pieces gives \vartheta(x) \le (2\ln 2)\,x, hence \psi(x) \le c_2\, x with Chebyshev's upper constant

Seeing the band

Here is \psi(x)/x plotted against x. It jitters — every prime power makes it jump — but it never escapes the horizontal corridor between Chebyshev's two constants c_1 \approx 0.921 (lower line) and c_2 \approx 1.106 (upper line). The dashed line at height 1 is where the Prime Number Theorem eventually says the wiggling ratio must settle — but Chebyshev's argument alone only guarantees the corridor, not the landing.

The picture makes the gap between Chebyshev and the PNT vivid: a band is not a limit. Watch how the ratio keeps crossing the dashed y=1 line back and forth. Chebyshev proved the walls of the corridor; showing that the wandering path actually converges to the centre took Riemann's zeta function and another four decades.

Worked example: an upper bound for π(x) from the binomial

Let's extract a concrete numerical bound on \pi(x) — the kind Chebyshev's method delivers — using only the central binomial coefficient. Recall

Every prime in (n, 2n] exceeds n, and there are \pi(2n) - \pi(n) of them, so their product is at least n^{\,\pi(2n)-\pi(n)}. Hence

So \pi(2n) - \pi(n) \le \dfrac{2\ln 2}{\ln n}\,n. Summing this over the dyadic blocks x, x/2, x/4, \dots telescopes to an upper bound of the shape

a completely explicit ceiling on the number of primes, obtained without a whisper of complex analysis. The matching floor comes from the lower bound on \binom{2n}{n} the same way. That both constants sit within about 40\% of 1 — and Chebyshev's sharper bookkeeping pulls them to within 10\% — is the whole force of the theorem.