9

In order to tackle this problem I first observed that

$$\phi(p_1^{e_1} \space p_2^{e_2} \cdots \space p_k^{e_k}) = (e_1 + 1)(e_2 + 1)\cdots(e_k +1)$$

Where $\phi(m)$ is the number of (not necessarily prime) divisors of $m$. If $m$ is the smallest integer such that $\phi(m) = n$, then

$$\phi(m) = n$$ $$(e_1 + 1)(e_2 + 1)\cdots(e_k +1) = n$$

Now we must choose $e_i$ such that $\prod_{i} p_i^{e_i}$ is minimal. The choices for $p$ are trivial - they are just the primes in ascending order.

However, my first thought for choosing $e_i$ was incorrect. I thought you could simply factor $n$, sort the factors in descending order and subtract 1. Most of the time this works fine, e.g. the smallest integer with $n = 15$ divisors is:

$$15 = 5 \cdot 3$$ $$15 = (4 + 1)(2 + 1)$$ $$m = 2^4 3^2 = 144$$

But this is incorrect for $n = 16$:

$$16 = 2 \cdot 2\cdot 2 \cdot 2$$ $$16 = (1 + 1)(1 + 1)(1 + 1)(1 + 1)$$ $$m = 2^1 3^1 5^1 7^1 = 210$$

Whereas the correct answer is:

$$16 = (3 + 1)(1 + 1)(1 + 1)$$ $$m = 2^3 3^1 5^1 = 120$$

So it's clear sometimes we need to merge factors. In this case because $7^1 > 2^2$. But I don't exactly see a clean and direct merging strategy. For example, one might think we must always merge into the $2$ power, but this is not true:

$$1552 = (96 + 1)(1 + 1)(1 + 1)(1 + 1)(1 + 1)$$ $$m = 2^{96} 3^1 5^1 7^1 11^1 > 2^{96} 3^3 5^1 7^1$$

I can't immediately think of an example, but my instinct says that some greedy approaches can fail if they merge the wrong powers first.

Is there a simple optimal strategy for merging these powers to get the correct answer?

Addendum. A greedy algorithm that checks every possible merge and performs the best one on a merge-by-merge basis, fails on $n = 3072$. The series of one-by-one merges is:

$$2^2 3^1 5^1 7^1 11^1 13^1 17^1 19^1 23^1 29^1 31^1$$

$$2^3 3^2 5^1 7^1 11^1 13^1 17^1 19^1 23^1 29^1$$

$$2^5 3^3 5^1 7^1 11^1 13^1 17^1 19^1 23^1$$

However the optimal solution is:

$$2^7 3^3 5^2 7^1 11^1 13^1 17^1 19^1$$

orlp
  • 13,988
  • 1
  • 26
  • 41

2 Answers2

1

Here's a solution, based on my comments above. I make no claims this is optimal.

The idea is to consider $T(n,m)$, which we defines as "the smallest positive integer with exactly $n$ divisors and $m$ distinct prime factors". We make the easy observations:

\begin{align} T(n, 1) &= 2^{n-1}\\ T(2^m, m) &= p_1p_2\cdots p_m \end{align}

And we also have the recurrence:

\begin{equation} T(n,m) = \min_{d|n} [T\left(\frac{n}{d}, m-1\right)\cdot p_m^{d-1}] \end{equation}

Finally, the quantity you're looking for is \begin{equation} \min_{1\le i\le\lceil\log(n)\rceil} T(n, i) \end{equation}

To that end, here is some Python code, which agrees with all the numbers you gave above. Note that it works with the logarithms to keep the numbers smaller: so the actual integer you seek is round(2**smallest(n)).

import functools
import itertools
import math

# All primes less than 100.
PRIMES = [
  2, 3, 5, 7, 11,
  13, 17, 19, 23, 29,
  31, 37, 41, 43, 47,
  53, 59, 61, 67, 71,
  73, 79, 83, 89, 97,
]

LOG_PRIMES = [math.log2(p) for p in PRIMES]

def smallest(n):
  max_factors = math.ceil(math.log2(n))
  min_so_far = float('Infinity')
  factors = factorize(n)
  memo = {}
  for i in range(1, max_factors+1):
    t = T(n,i, factors, memo)
    if 0.0 < t < min_so_far:
      min_so_far = t
  return min_so_far

def T(n, m, factors=None, memo=None):
  if memo is None:
    memo = {}
  if n < 2 or m < 1:
    return 0
  elif m == 1:
    # Everything on the smallest prime.
    return (n-1) * LOG_PRIMES[0]
  elif n < 2**m:
    return 0
  elif n == 2**m:
    # Product of first m primes, in log.
    return sum(LOG_PRIMES[:m])
  elif (n,m) in memo:
    return memo[(n,m)]

  if factors is None:
    factors = factorize(n)
  if len(factors) < m:
    return 0

  smallest = float('Infinity')  
  for factor_list in powerset(factors):
    divisor = product(factor_list)
    first = T(divisor, m-1, factor_list, memo)
    # No such product.
    if first < 1.0:
      continue
    second = (n/divisor - 1) * LOG_PRIMES[m-1]
    total = first + second
    if total < smallest:
      smallest = total

  memo[(n,m)] = smallest
  return smallest

def product(nums):
  return functools.reduce(lambda x,y: x*y, nums, 1)

def factorize(n):
  prime_factors = []
  for p in PRIMES:
    while n%p == 0:
      n //= p
      prime_factors.append(p)
    if n == 1:
      break
  return prime_factors

def powerset(lst):
  # No empty set.
  return itertools.chain.from_iterable(itertools.combinations(lst, r) 
                                       for r in range(1, len(lst)+1))
Steve D
  • 166
  • 5
-1

Possible candidates for "smallest integer with n divisors" are the integers of the form $2^a·3^b·5^c...$ where a ≥ b ≥ c ... and (a+1)(b+1)(c+1)... = n.

So you need to find all ways to express n as the product of integers ≥ 2 in non-ascending order, and calculate and check the corresponding candidates. For example, when n = 16, 16 = 8·2 = 4·4 = 4·2·2 = 2·2·2·2, so possibilities are $2^7·3$, $2^3·3^3$, $2^3·3·5$, $2·3·5·7$, and the smallest is $2^3·3·5 = 120$.

If n is the product of two primes p·q, p ≥ q, the only candidates are $2^{pq-1}$ and $2^{p-1}·3^{q-1}$, and the latter is always smaller.

You can figure out some condition when there could be a factor $2^{ab-1}$ for example, by checking whether $2^{ab-1} > 2^{a-1}·x^{b-1}$ for some prime x that isn't a factor. In the example n = 16, there is a factor $2^3$ because $2^3 < 2·7$.

gnasher729
  • 32,238
  • 36
  • 56