7

Let $x$ be a rational number, and $S_n(x)= \sum_{1\leq i\leq n} i^x$. What is the complexity of computing $S_n(x)$ correct to $d$ decimal places? Is this a Hard problem?

It is clear from Faulhaber's formula that the problem takes $O(x)$ steps to solve for natural $x$. But I'm not very clear about rational $x$.

Ganesh
  • 521
  • 3
  • 9
  • 4
    related: the sum of square roots problem (i.e a general version of S_n(1/2)): http://cstheory.stackexchange.com/questions/79/problems-between-p-and-npc/4010#4010 – Suresh Venkat Jan 15 '11 at 22:31

2 Answers2

9

Check out The generalization of Faulhaber's formula to sums of non-integral powers. Specifically, on the last page, it gives a formula indicating an error bound of $O(N^{\sigma - m})$, where $\sigma$ is $x$ (in this paper, it's generalized to complex numbers), and $m = \lfloor \sigma + 1 \rfloor$. $N$ in this case refers to the depth of a summation you have to calculate. So this implies that for rational numbers slightly greater than an integer, it would be a lot faster than for rational numbers slightly less than an integer. Of course, this is a result on real numbers, and there may be a better result for rationals.

Gautam Kamath
  • 436
  • 3
  • 5
2

Perhaps the Euler-Maclaurin formula can help.

Yuval Filmus
  • 14,445
  • 1
  • 48
  • 92