Given two natural numbers $i$ and $p$ such that $0 < i \leqslant 2^p$, let $$ \psi(p,i) := p - \alpha + 1 - \frac{1}{2^p}\left((2^p+i)\lg(2^p+i) - i\lg i - i + \alpha - \frac{2^p}{i+1} - \frac{i}{2^p+1}\right), $$ where $\alpha \simeq 1.264499$ and $\lg n$ is the binary logarithm of $n$. I am looking for $\min_{p,i}\psi(p,i)$.

It seems that $\min_{i}\psi(p,i) = \psi(p,2^p)$ or $\psi(p,2^p-1)$. Anyway, I can then minimize over $p$.

We have $$ \frac{\partial\psi}{\partial i}(p,i) = - \frac{1}{(i+1)^2} - \frac{1}{2^p}\left(\lg(2^p+i) - \lg i - \frac{1}{2^p+1} - 1\right). $$ (The derivative with respect to $p$ is much worse.) Also $$ \lim_{i \rightarrow 0^{+}}\frac{\partial\psi}{\partial i}(p,i) = -\infty,\quad \text{and}\quad \left.\frac{\partial\psi}{\partial i}(p,i)\right|_{i=2^p} \!\!= \frac{1}{2^p(2^p+1)^2} > 0. $$ The derivative is strictly increasing (the second derivative is positive). How can I prove that the root of $\partial\psi/\partial i = 0$ is between $i=2^p-1$ and $i=2^p$? And which of these values is the minimum?

  • $\begingroup$ Something can't be right in your minimum formula. You take the minimum over $p$ and $i$ and the result still contains $p$? $\endgroup$ – Raskolnikov Sep 5 '12 at 10:19
  • $\begingroup$ @Raskolnikov: You are right. Fixed. $\endgroup$ – Christian Sep 5 '12 at 10:27

It is tedious, even with the help of some computer algebra system, to check that $\left.\partial\psi/\partial i\right|_{i=2^p-1} < 0$, so the root is indeed between $2^p-1$ and $2^p$. It is then fastidious to check that $\psi(p,2^p-1) > \psi(p,2^p)$, so $\min_{0<i\leqslant 2^p}\psi(p,i) = \psi(p,2^p)$.

After that, the minimization of $\psi(p,2^p)$ yields $\min_{0 < p, 0 < i \leqslant 2^p}\psi(p,i) = -\alpha$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.