I managed to find an interesting inequality containing the sum of the divisors of a number and the number of them, using AM-GM between each 2 of the divisors of it: $\sigma(n)\geq\sqrt n.d(n)+(\sqrt n-1)^2$. I tried to use this to find a bound for the average of divisors of $n$, and I got: $\frac{\sigma(n)}{d(n)}\geq\sqrt n+\frac{n-2\sqrt n+1}{d(n)}$(I) and we know that $d(n)\leq2\sqrt n$ so $\frac{1}{d(n)}\geq\frac{1}{2\sqrt n}$. Using this on inequality (I) we get: $\frac{\sigma(n)}{d(n)}\geq\sqrt n+\frac{n-2\sqrt n+1}{d(n)}\geq\sqrt n+\frac{n-2\sqrt n+1}{2\sqrt n}=\frac{3}{2}\sqrt{n}-1+\frac{1}{2\sqrt n}\geq\frac{3}{2}\sqrt{n}-1$
Now we have a lower bound for $\frac{\sigma(n)}{d(n)}$. According to this question, we can also say $\frac{\sigma(n)}{d(n)}\leq\frac{3}{8}n+1$ for composite $n>4$.
To sum it up, for composite $n>4$ have $\frac{3}{8}n+1\geq\frac{\sigma(n)}{d(n)}\geq\frac{3}{2}\sqrt{n}-1$
Can this bound be improved? I realized the higher bound is very good, but the lower bound will not be good for large $n$. I would appreciate any help.
2026-03-26 08:00:35.1774512035
Bounds on the average of the divisors of natural numbers.
680 Views Asked by Bumbble Comm https://math.techqa.club/user/bumbble-comm/detail At
1
$$ \frac{\sigma(n)}{d(n)} \; \geq \; \frac{14}{3} \; \left(\frac{n}{12} \right)^{\left( \frac{2}{3} \right)} $$
$$ \frac{\sigma(n)}{d(n)} \; \geq \; \frac{9762}{5} \; \left(\frac{n}{55440} \right)^{\left( \frac{3}{4} \right)} $$
$$ \color{magenta}{ \frac{\sigma(n)}{d(n)} \; \geq \; 340156800 \; \left(\frac{n}{321253732800} \right)^{\left( \frac{4}{5} \right)} } $$
These are fairly difficult. The procedure was invented by Ramanujan and is what defines the superior highly composite numbers and the colossally abundant numbers. There was a bit of a trick needed here because of nontrivial number theoretic multiplicative functions in both numerator and denominator.
I am going to submit the following sequence to the OEIS as the Miserable Average Divisor numbers. Given real $\delta > 0,$ such a number minimizes $$ \frac{n^\delta \sigma(n)}{n d(n)} = \frac{ \sigma(n)}{n^{1-\delta} \, d(n)}. $$ Let's see, for a fixed prime $p$ and exponent $k,$ the first (largest) $\delta$ that causes $p$ to get exponent $k$ is $$ \delta = \frac{\log(k+1) + \log p + \log \left(p^k -1 \right) - \log k - \log \left(p^{k+1} -1 \right) }{\log p}. $$ This is enough information to find the first MAD numbers by placing some bound on $\delta$ and ordering them. Now, the other thing we really want is to solve for $k$ once given a prime $p$ and a real $\delta.$ So far, I have been unable to get a closed form for $k$ using the floor function, which is how it usually works. Can't win them all.