Distribution Of Divisor Sums Modulo 32

For each integer 2 k n 2 \le k \le n , choose a divisor d k d_k of k k , uniformly at random from the set of divisors of k . k. We denote by P ( n ) P(n) the probability that

d 2 + d 3 + + d n d_2 + d_3 + \cdots + d_n

is divisible by 32.

Amazingly, there exists a positive integer N N such that for all n N n\ge N , the value of P ( n ) P(n) is exactly 1 32 \frac1{32} . What is the smallest N N for which this is true?


The answer is 27.

This section requires Javascript.
You are seeing this because something didn't load right. We suggest you, (a) try refreshing the page, (b) enabling javascript if it is disabled on your browser and, finally, (c) loading the non-javascript version of this page . We're sorry about the hassle.

1 solution

Sameer Kailasa
Mar 30, 2016

Consider the polynomial p n ( x ) = k = 2 n ( d n x d ) = ( x + x 2 ) ( x + x 3 ) ( x + x 2 + x 4 ) p_n (x) = \prod_{k=2}^{n} \left(\sum_{d\vert n} x^d \right) = (x+x^2)(x+x^3)(x+x^2+x^4) \cdots Let ω \omega be a primitive 32 32 nd root of unity. Then, a roots of unity filter gives P ( n ) = 1 32 p n ( 1 ) i = 0 31 p n ( ω i ) . P(n) = \frac{1}{32 p_n (1)}\sum_{i=0}^{31} p_n (\omega^i). Now, note that for each i i , by Dirichlet's theorem , we can find a prime p p of the form 2 i + 1 m + 2 i + 1 = 2 i ( 2 m + 1 ) + 1 2^{i+1} m + 2^i + 1 = 2^i (2m+1)+1 so that x 2 i + 1 x^{2^i} + 1 divides x p + x x^p + x . In particular, there exists n 0 n_0 such that for all n n 0 n\ge n_0 , ( x 16 + 1 ) ( x 8 + 1 ) ( x 4 + 1 ) ( x 2 + 1 ) ( x + 1 ) p n ( x ) (x^{16} + 1)(x^{8} + 1)(x^4 + 1)(x^2 + 1)(x+1) \, \big\vert \, p_n (x) and hence p n ( ω ) = p n ( ω 2 ) = p n ( ω 31 ) = 0 p_n (\omega) = p_n (\omega^2) = \cdots p_n (\omega^{31}) = 0 . Accordingly, for n n 0 n\ge n_0 , the value of P ( n ) P(n) is exactly 1 / 32 1/32 .

Let q k ( x ) = d k x d q_k (x) = \sum_{d\vert k} x^d . To compute n 0 n_0 , for each 0 i 4 0\le i\le 4 we must find the smallest k k such that x 2 i + 1 x^{2^i} + 1 divides q k ( x ) q_k (x) . The above argument shows that some such k k exists for all i i , but the prime produced by Dirichlet's theorem is not necessarily the minimal such k k . From here, we factor q k q_k for small values of k k and find the minimal k k for each i i : x + 1 q 2 ( x ) x+1 \, \big\vert \, q_2 (x) x 2 + 1 q 3 ( x ) x^2 + 1 \big\vert \, q_3 (x) x 4 + 1 q 5 ( x ) x^4 + 1 \big\vert \, q_5 (x) x 8 + 1 q 27 ( x ) x^8 + 1 \big\vert \, q_{27} (x) x 16 + 1 q 17 ( x ) x^{16} + 1 \big\vert \, q_{17} (x) Taking the maximum of these indices, we conclude n 0 = 27 n_0 = 27 .

I posted a report, but I also had a separate comment: it's technically possible that the constant term of p n ( x ) p_n(x) mod x 32 1 x^{32}-1 is 1 / 32 1/32 without p n ( x ) p_n(x) being divisible by ( x 16 + 1 ) ( x 8 + 1 ) ( x 4 + 1 ) ( x 2 + 1 ) ( x + 1 ) . (x^{16}+1)(x^8+1)(x^4+1)(x^2+1)(x+1). (Like, if this happened to be the case for n = 26 , n=26, the answer would be 26. \le 26. ) I don't think there's any better way to check that this doesn't happen than to write down the first 26 values of p n ( x ) . p_n(x).

Patrick Corn - 2 years, 10 months ago

It actually does not happen - I checked it via maxima:

/* maxima program to find the lowest n */
P : x - 1$              /* P = p_n(x) * (x-1) mod (x^32 - 1) */
Q : x^32 - 1$

for n : 1 thru 100 do (
    factor : lsum(x^d,
        d, listify(divisors(n))
    ),

    P : remainder(P * factor, Q),

    R : diff(P, x),    /* P(n) = 1/32 iff R = 0 */
    display(R : ev(R, x = 1) + 32 * ev(P, x = 0)),

    if P = 0 then return(n)
);

We get the probability of 1 32 \frac{1}{32} if (and only if) R = P ( 1 ) ( 1 ) + 32 P ( 0 ) = 0 R = P^{(1)}(1) + 32P(0)=0 , which is never the case for n 26 n\leq 26 .

Carsten Meyer - 2 months ago

1 pending report

Vote up reports you agree with

×

Problem Loading...

Note Loading...

Set Loading...