I am teaching a class in linear algebra and I asked myself the following question: what is the chance to get an invertible matrix if I write a random one? My impulsive answer is "very likely". Of course, over the reals the probability is $1$ because singular matrices have codimension 1.
On the other hand, if we restrict to integer numbers and bound the maximum coefficient $N$ that appears, the answer becomes much more difficult ( see here). I then recalled there is a discrete case in which one manages to calculate a number. Indeed, over finite fields the fraction of $\mathbb{F}_q$- invertible matrices is (see here)
$$ \left ( 1- \frac{1}{q^n} \right ) \cdot \ldots \cdot \left ( 1- \frac{1}{q} \right ) \simeq 1- \frac{1}{q} = \frac{ q-1}{q} $$
This however excludes the matrices which have determinant multiple of $q$ but nonzero. We conclude that there are at least $(q-1)/q$ invertible matrices with nonnegative coefficients $< q$. My question is
Can you generalize this estimate for non prime $q$ ?