8

I was thinking about derivative of infinite sum of functions, i.e.

$$f(x) = \sum_{i = 0}^\infty g_i(x)$$

$g(x)$ is continuous in domain of $f$

Because if $(f+g)'(x) = f'(x) + g'(x)$ then $\left(\sum\limits_{i = 0}^{\infty} g_i(x)\right)' = \sum\limits_{i = 0}^{\infty} g_i'(x)$ isn't it?

Hauleth
  • 274

1 Answers1

10

First I assume you mean $g_i$ instead of $g$, and you have to suppose at least that the $g_i$ are all differentiable (more than just continuous).

Even then this is in general false. One common case where it is true is when you assume uniform convergence of $\sum g_i^{'}$ and at least one point of convergence for $\sum g_i$.

A counter example under your hypothesis : take $g_i^{'}(x) = \cos(i \pi x)/i^2$. then $\sum g_i$ converges since it converges normally ($\sum \frac{1}{i^2}< \infty$) but $\sum g_i^{'}$ diverges at 0 (since $\sum \frac{1}{i} = \infty$).

Michael Hardy
  • 1
Albert
  • 9,170
  • And what if $g_i(x) = 1/i^x$? That is especially interesting me. – Hauleth Nov 24 '11 at 18:46
  • then it is true for $x>1$ because the convergence is uniform on all compacts of $]1, +\infty[$ – Albert Nov 24 '11 at 18:53
  • $\sum_{i=1}^\infty 1/i^x$ is the Riemann zeta function $\zeta(x)$ for $\Re x > 1$. The series of derivatives $\sum_{i=1}^\infty -\ln(i)/i^x$ also converges for $\Re x > 1$, and uniformly on compact sets, so by the "One common case" Glougloubarbaki mentioned the sum is indeed $\zeta'(x)$ for $\Re x > 1$. – Robert Israel Nov 24 '11 at 18:54