I have a problem to solve. N natural number is given. I need to find a list of natural numbers which sum up to that given number and at the same time the inverses up to 1.
a + b + c + ... = N
1/a + 1/b + 1/c + ... = 1
a, b, c don't have to be unique.
I have come up with following code in Java. It works for simple cases, but incredibly slow for already for N > 1000.
How can I rewrite the method so it works fast even for millions? Maybe, I should drop off recursion or cut off some of branches with mathematical trick which I miss?
SSCEE:
private final static double ONE = 1.00000001;
public List<Integer> search (int number) {
    int bound = (int)Math.sqrt(number) + 1;
    List<Integer> list = new ArrayList<Integer>(bound);
    if (number == 1) {
        list.add(1);
        return list;
    }
    for (int i = 2; i <= bound; i++) {
        list.clear();
        if (simulate(number, i, list, 0.0)) break;
    }
    return list;
}
//TODO: how to reuse already calculated results?
private boolean search (int number, int n, List<Integer> list, double sum) {
    if (sum > ONE) {
        return false;
    }
    //would be larger anyway
    double minSum = sum + 1.0 / number;
    if (minSum > ONE) {
        return false;
    }
    if (n == 1) {
        if (minSum < 0.99999999) {
            return false;
        }
        list.add(number);
        return true;
    }
    boolean success = false;
    for (int i = 2; i < number; i++) {
        if (number - i > 0) {
            double tmpSum = sum + 1.0 / i;
            if (tmpSum > ONE) continue;
            list.add(i);
            success = search(number - i, n - 1, list, tmpSum);
            if (!success) {
                list.remove(list.size() - 1);
            }
            if (success) break;
        }
    }
    return success;
}