Probabilistic lower bounds on maximal determinants of binary matrices (original) (raw)

View PDF

Abstract:Let mathcalD(n){\mathcal D}(n)mathcalD(n) be the maximal determinant for ntimesnn \times nntimesn pm1\{\pm 1\}pm1-matrices, and mathcalR(n)=mathcalD(n)/nn/2\mathcal R(n) = {\mathcal D}(n)/n^{n/2}mathcalR(n)=mathcalD(n)/nn/2 be the ratio of mathcalD(n){\mathcal D}(n)mathcalD(n) to the Hadamard upper bound. Using the probabilistic method, we prove new lower bounds on mathcalD(n){\mathcal D}(n)mathcalD(n) and mathcalR(n)\mathcal R(n)mathcalR(n) in terms of d=n−hd = n-hd=n−h, where hhh is the order of a Hadamard matrix and hhh is maximal subject to hlenh \le nhlen. For example, mathcalR(n)>(pie/2)−d/2\mathcal R(n) > (\pi e/2)^{-d/2}mathcalR(n)>(pie/2)−d/2 if 1ledle31 \le d \le 31ledle3, and mathcalR(n)>(pie/2)−d/2(1−d2(pi/(2h))1/2)\mathcal R(n) > (\pi e/2)^{-d/2}(1 - d^2(\pi/(2h))^{1/2})mathcalR(n)>(pie/2)−d/2(1−d2(pi/(2h))1/2) if d>3d > 3d>3. By a recent result of Livinskyi, d2/h1/2to0d^2/h^{1/2} \to 0d2/h1/2to0 as ntoinftyn \to \inftyntoinfty, so the second bound is close to (pie/2)−d/2(\pi e/2)^{-d/2}(pie/2)−d/2 for large nnn. Previous lower bounds tended to zero as ntoinftyn \to \inftyntoinfty with ddd fixed, except in the cases din0,1d \in \{0,1\}din0,1. For dge2d \ge 2dge2, our bounds are better for all sufficiently large nnn. If the Hadamard conjecture is true, then dle3d \le 3dle3, so the first bound above shows that mathcalR(n)\mathcal R(n)mathcalR(n) is bounded below by a positive constant (pie/2)−3/2>0.1133(\pi e/2)^{-3/2} > 0.1133(pie/2)−3/2>0.1133.

Submission history

From: Richard Brent [view email]
[v1] Mon, 26 Jan 2015 02:19:57 UTC (13 KB)
[v2] Tue, 17 Feb 2015 23:22:14 UTC (13 KB)
[v3] Sun, 1 Mar 2015 00:32:24 UTC (13 KB)
[v4] Tue, 26 Jan 2016 10:55:56 UTC (14 KB)
[v5] Mon, 1 Aug 2016 00:40:38 UTC (14 KB)
[v6] Thu, 18 Aug 2016 06:10:20 UTC (15 KB)
[v7] Tue, 25 Oct 2016 04:16:48 UTC (15 KB)