In a paper about color-coding they estimate the complexity of coloring trials needed to achieve an $\varepsilon$-error for $k\in \mathbb{N}$:
$$ \left\lceil\frac{ln(\varepsilon)}{ln{(1-k!/k^k)}} \right\rceil = |ln(\varepsilon)| \cdot \mathcal{O}(e^k)$$
For me it is clear, why there is the $ln(\varepsilon)$ term, but I do not understand how the $\mathcal{O}(e^k)$ is calculated. I think, one should show, that following holds:
$$ \left| \frac{1}{ln(1-k!/k^k)} \right| \in \mathcal{O}(e^k) $$ A bound $\frac{k!}{k^k} > e^{-k}$ was stated in the paper, so it follow:
$$ \left| \frac{1}{ln(1-k!/k^k)} \right| < \left| \frac{1}{ln(1-e^{-k})} \right| $$
Therefore, the statement can be shown, by proving that:
$$ \left| \frac{1}{ln(1-e^{-k})} \right| \in \mathcal{O}(e^k) $$
But at this point i am stuck. My math knowledge is a little bit rusted unfortunately.