Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Let $2\leq d_1 < d_2,...,d_l < n$ be all the proper nontrivial divisors of $n$. I like to understand how much these divisors deviates from each other. Here are two questions in this regard:

(1) What is the maximum of the set $\{d_i/d_{i-1}: 1\leq i \leq l\}$. Say it $M$.

Assume that you know the prime factorization of $n (= p_1^{\alpha_1}..p_r^{\alpha_r})$. Can I have a formula in terms of $p_i$'s and $\alpha_i$'s. One of the crude upper bound can be $p_r$ but this is really bad if $n$ has many distinct prime factors.

(2) Instead of maximum, the average may be more interesting and useful. So can we estimate the mean and variance,

$\mu = (1/l)\sum_{i=1}^{l}{d_i/d_{i-1}}$ and $\sigma^2 = (1/l)\sum_{i=1}^{l}{(d_i/d_{i-1}-\mu)^2}$

share|improve this question
    
I would ask Kevin Ford at the University of Illinois. A few years he published a long paper in the Annals about divisors in an interval. –  M Khan Dec 7 '12 at 13:55
1  
    
But $H(x, y, z) - H(x-\Delta, y, z)$ can be computed only if $\Delta \geq x/\log^{10}{z}$, (as in Introduction, Theorem 2, in above mentioned paper). Then recovering the integer whose divisor we have to look at is not possible by this formula. –  Kamalakshya Dec 7 '12 at 16:37
    
This may be worth studying if n has many divisors. What significance do you see it having for n having fewer than 5 divisors? Or 10? Gerhard "Ask Me About System Design" Paseman, 2012.12.07 –  Gerhard Paseman Dec 7 '12 at 17:11
    
Also, assuming d_0 is 1, it is clear that M is at least p_1, where I assume the prime factors are indexed in order of increasing magnitude, and that for n with many factors, log of the mean will be close to if not equal to log of n divided by number of divisors of n. There may be some bizarre exceptions like a twice a large ppprime taken to a large power, but they should be easy to characterize. Gerhard "Ask Me About Gut Feeling" Paseman, 2012.12.07 –  Gerhard Paseman Dec 7 '12 at 17:21
show 6 more comments

1 Answer

up vote 5 down vote accepted

Tenenbaum [Sur un probleme de crible et ses applications (1986)] showed that $$ F(n)/n=\max_{1\le i < \tau(n)} \frac{d_{i+1}(n)}{d_i(n)},$$ where $1=d_1(n)< \ldots < d_{\tau(n)}(n)=n$ is the increasing sequence of divisors, and $F$ is given by $F(1)=1$ and $F(n)=\max \{ d P^{-}(d) : d|n,\\, d>1 \}$ for $n\ge 2$, where $P^{-}(n)$ is the smallest prime divisor of $n$. For $n=p_1^{\alpha_1}\cdots p_k^{\alpha_k}$, with $p_1<\ldots < p_k$, this leads to the formula $$ \max_{1\le i < \tau(n)} \frac{d_{i+1}(n)}{d_i(n)} = \max_{1 \le j \le k} \{ p_j p_j^{\alpha_j}\cdots p_k^{\alpha_k} \}/n$$

Let $D(x,t)=|\{n\le x: F(n)/n \le t\}|$. I showed [Integers with dense divisors, 2 (2004)] that $D(x,t)=x\\, d(w) \{1+O(1/\log t) \}$, for $x\ge 3$, $x\ge t \ge \exp\{(\log(\log(x)))^{5/3+\varepsilon}\}$, where $w=\log x / \log t$ and $d(w)$ is a continuous function that can be expressed in terms of Dickman's function and which satisfies $d(w) \asymp 1/w$.

share|improve this answer
    
In definition of $F(n)$ is $d\leq n$ ? –  Kamalakshya Dec 8 '12 at 2:11
    
Yes, $d|n$ implies $d\le n$ in the definition of $F$. –  Andreas Weingartner Dec 8 '12 at 2:17
    
small typo $\max_{1\le i < \tau(n)} \frac{d_{i+1}(n)}{d_i(n)} = \max_{\substack{0 < \beta_j \le \alpha_j \\ 0 \le \beta_{j+l} \le \alpha_{j+l}\\ 1 \le l}} \{ p_j p_j^{\beta_j}\cdots p_k^{\beta_k} \}/n$ –  Kamalakshya Dec 8 '12 at 4:30
    
I don't think there is a typo. Because you are looking for a maximum, you can take $\beta_i = \alpha_i$. –  Andreas Weingartner Dec 8 '12 at 4:40
    
yes, what I was thinking is reductive. –  Kamalakshya Dec 8 '12 at 4:45
show 2 more comments

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.