Wiki
Wiki

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

Updated

Erdos 1946 asymptotic number latin rectangles

../

series_p234: Erdős and Kaplansky's sketch, without full proof, that the error in their Theorem 1 is of order k^2/n, and that f(n,k) (n!)^{-k} exp(binom(k,2)) has the expansion 1 - binom(k,3)/n + binom(k,3)(k^3 - 3k^2 + 8k - 30)/(12n^2) + ..., which they remark suggests the formula fails at about k = n^{1/3}.

theorem_1: Erdős and Kaplansky's one-row step: if k < (log n)^{3/2-ε}, then for sufficiently large n the number N of ways to add a (k+1)-st row to a k-row Latin rectangle on the symbols 1, ..., n satisfies |N e^k/n! - 1| < n^{-c}, with c > 0 depending only on ε.

theorem_2: Erdős and Kaplansky's main result: if f(n,k) is the number of n by k Latin rectangles (k rows on the symbols 1, ..., n) and k < (log n)^{3/2-ε}, then f(n,k) (n!)^{-k} exp(binom(k,2)) tends to 1 as n tends to infinity.


P. Erdős, I. Kaplansky: The asymptotic number of Latin rectangles, Amer. J. Math. 68 (1946) no. 2, 230--236 (MR 7,407b; Zentralblatt 60,28), DOI 10.2307/2371834. No notice is printed in the file (the scan's first and last pages carry no copyright or license line); the hosting archive's site footer speaks for the site, not the paper (https://users.renyi.hu/~p_erdos/, read 2026-10-02, prints "(C) 2005-2007 All rights reserved. All material on this site is for scientifics purposes only."); the publisher's page for the 1946 article was not consulted for this file, and the Crossref record names no license; the term is unstated.

The paper proves the asymptotic formula for the number f(n,k) of n by k Latin rectangles that the introduction calls an easy heuristic conjecture (p. 230). An n by k Latin rectangle here has k rows, each an arrangement of 1, ..., n, with distinct entries in each column; the printed definition (p. 230) says n rows and k columns but puts 1, ..., n in each row and adds rows one at a time. Theorem 1 (p. 232) shows that if k < (log n)^{3/2-eps}, then for sufficiently large n the number N of ways to add a (k+1)-st row to a k-row rectangle satisfies |N e^k/n! - 1| < n^{-c}, where c > 0 depends only on eps. Theorem 2 (p. 234) multiplies these row-by-row estimates to give f(n,k) (n!)^{-k} exp(binom(k,2)) -> 1 as n tends to infinity, in the same range of k. The method is a double application of inclusion-exclusion: first over columns where a candidate new row clashes with the rectangle, then over pairs of equal entries, giving the auxiliary counts A_r and B(r,s) that are then estimated. Section 4 (pp. 234--236) sketches, without full proofs, further terms of the asymptotic series, (20) for N and (21) for f(n,k). The introduction (p. 230) says (log n)^{3/2} "appears to be a 'natural boundary' of the method" and that the authors believe the actual break occurs at k = n^{1/3}; p. 236 adds that the form of (21) suggests the formula stops holding at about k = n^{1/3}, which the authors cannot prove.

Source: https://users.renyi.hu/~p_erdos/1946-12.pdf.

Bears on.

  • #725: the problem asks for an asymptotic formula for the number of k by n Latin rectangles without restricting k; Theorem 2 gives f(n,k) ~ (n!)^k exp(-binom(k,2)) for k < (log n)^{3/2-eps} and says nothing about larger k. Section 4 sketches, without proof, further terms of the expansion to order n^{-2} and records the authors' unproved expectation that the formula fails near k = n^{1/3}.

Results.

  • Theorem 1 (p. 232): for k < (log n)^{3/2-eps} and n large, a k-row Latin rectangle on 1, ..., n has N extensions by one row with |N e^k/n! - 1| < n^{-c}, c > 0 depending only on eps.
  • Theorem 2 (p. 234), the main result: f(n,k) (n!)^{-k} exp(binom(k,2)) -> 1 as n -> infinity when k < (log n)^{3/2-eps}.
  • Section 4 (pp. 234--236): the sketched expansions (20) and (21), with the error in Theorem 1 of order k^2/n and the remark that the formula appears to fail at about k = n^{1/3}.

Read status. Claims checked: the three results above were read clause by clause on the page images of the print; Section 4 is a sketch in the paper and is recorded as one. Nothing is independently reviewed.

No file of this source is held: no license on record permits its redistribution, and the card cites the edition it names above.