Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Statement

Let nn and kk be the integers of a binomial coefficient (nk)\binom nk, and let pp be a prime. If pap^a divides (nk)\binom nk, then

pa≤n.p^a\le n.

Equivalently, if aa is defined by pa≤n<pa+1p^a\le n<p^{a+1}, then the exponent of pp in (nk)\binom nk is at most aa; this is the form in which the print concludes its proof. The print states the lemma with no range on nn and kk; the theorem applies it in its binomial form, with n≥2kn\ge2k.

Source. P. Erdős, A theorem of Sylvester and Schur, J. London Math. Soc. 9 (1934), no. 4, 282--288, DOI 10.1112/jlms/s1-9.4.282; the unnumbered lemma and its proof on printed p. 283 (PDF p. 2 of the seven-page offprint scan, read on the page image). The source card is [[factorials_binomials/erdos_1934_theorem_sylvester_schur/_index|A theorem of Sylvester and Schur]].

Read depth. Claims checked: the statement was read clause by clause on the page image. The short proof was read and is sketched below; it is not independently verified.

Proof pointer

By Legendre's formula the exponent of pp in n!/((n−k)! k!)n!/((n-k)!\,k!) is the sum over i≥1i\ge1 of [n/pi]−[k/pi]−[(n−k)/pi][n/p^i]-[k/p^i]-[(n-k)/p^i]. Each such term is 00 or 11, since [x+y]−[x]−[y]∈{0,1}[x+y]-[x]-[y]\in\{0,1\}, and the terms with pi>np^i>n vanish. So at most aa terms are nonzero, where pa≤n<pa+1p^a\le n<p^{a+1}.

Dependencies

Legendre's formula for the exponent of a prime in a factorial. Nothing else.

Use in the paper

Under the hypothesis that (nk)\binom nk has no prime factor greater than kk, the lemma bounds the coefficient by nπ(k)n^{\pi(k)}; with π(k)≤k/2\pi(k)\le k/2 for k≥8k\ge8 this settles the theorem for 8≤k≤n8\le k\le\sqrt n (p. 283), and with π(k)<k/3\pi(k)<k/3 for k>37k>37 for 37<k≤n2/337<k\le n^{2/3} (p. 284). For k>n2/3k>n^{2/3} and k>37k>37 it gives the bound of the coefficient by nested products of primes up to kk, n\sqrt n, n3\sqrt[3]n, and so on (p. 284).

Bears on

  • Problem 961: the lemma is a step of Erdős's proof (pp. 283--288) of the theorem, which in that problem's notation is the bound f(k)≤kf(k)\le k; the lemma alone gives no bound on f(k)f(k).