Abstract: This papers contains two results concerning random $n \times n$ Bernoulli matrices. First, we show that with probability tending to one the determinant has absolute value $\sqrt {n!} \exp(O(\sqrt(n log n)))$. Next, we prove a new upper bound $.939^n$ on the probability that the matrix is singular. We also give some generalizations to other random matrix models.
What problem does this paper attempt to address?
This paper attempts to solve two main problems regarding random \( \pm 1 \) matrices:
### Problem 1: What is the typical value of the determinant of a random matrix?
Specifically, given an \( n\times n \) random Bernoulli matrix \( M_n \), whose elements are independent and identically distributed (i.i.d.) Bernoulli random variables (that is, each element takes \( + 1\) or \( - 1\) with equal probability). The paper focuses on the typical magnitude of the absolute value of the determinant \( \det(M_n)\) of this matrix.
According to the Hadamard inequality, we have:
\[|\det(M_n)|\leq n^{n/2}\]
The equality holds if and only if \( M_n \) is a Hadamard matrix. However, in general, we expect \( |\det(M_n)|\) to be somewhat smaller than \( n^{n/2}\). Through simple estimation:
\[E((\det M_n)^2)=n!\]
It can be conjectured that \( |\det(M_n)|\) should be approximately:
\[\sqrt{n!}=e^{-n/2 + o(n)}n^{n/2}\]
The first main result of the paper shows that as \( n \) tends to infinity, the absolute value of the determinant is very close to \( \sqrt{n!}\):
\[P(|\det M_n|\geq\sqrt{n!}\exp(- 29n^{1/2}\ln^{1/2}n)) = 1 - o(1)\]
### Problem 2: What is the probability that a random matrix is singular?
Another problem is to calculate the probability that \( M_n \) is singular (that is, the determinant is zero). Komlós proved in 1967 that:
\[P(\det M_n = 0)=o(1)\]
That is, as \( n \) increases, the probability that the matrix is singular tends to zero.
More precisely, the conjecture is that the main source of singularity is that there are two rows or two columns in the matrix that are exactly the same or opposite, so:
\[P(\det M_n = 0)\approx(1 - o(1))n^22^{1 - n}\]
The best previous partial result was given by Kahn, Komlós and Szemerédi:
\[P(\det M_n = 0)\leq(1-\epsilon+o(1))^n\]
where \( \epsilon := 0.001\).
The second main result of the paper improves this conclusion:
\[P(\det M_n = 0)\leq(1-\epsilon+o(1))^n\]
where \( \epsilon := 0.06191\ldots\), which is the unique solution of the equation:
\[h(\epsilon)+\frac{\epsilon}{\log_2\frac{16}{15}} = 1\]
in the interval \( (0,1/2)\), and \( h(\epsilon)\) is the entropy function:
\[h(\epsilon):=\epsilon\log_2\frac{1}{\epsilon}+(1-\epsilon)\log_2\frac{1}{1-\epsilon}\]
In summary, this paper solves two key problems regarding the typical value of the determinant of a random \( \pm 1\) matrix and the probability of matrix singularity, and provides more accurate results.