Let $0<a_1,\dots,a_n<1\leq b_1,\dots,b_n$. For $1\leq i\leq n$, define $$M_i=\frac{1}{\dfrac{b_i-a_i}{b_i-a_1}+\dfrac{b_i-a_i}{b_i-a_2}+\dots+\dfrac{b_i-a_i}{b_i-a_n}}.$$ Let $M=M_1+\dots+M_n$. Is $M$ bounded above by some constant (independent of $n,a_1,\dots,a_n,b_1,\dots,b_n$)?

If we assume (without loss of generality) that $a_1\geq\dots\geq a_n$, we get $M_i\leq\frac{1}{i}$, so $M\leq 1+\frac12+\dots+\frac1n$. But this sum is not bounded by a constant.

  • $\begingroup$ $n$ is fixed in each case and the upper bound should be intended about quantities $a_i$ and $b_i$ for $i=1,2,...,n$ $\endgroup$ – Piquito Sep 1 '16 at 14:42

A tough nut to crack. But the upper bound you give is in fact optimal. So there is no finite upper bound, uniform in $n$:

Fix $n$, let $\Lambda>1$ and pick numbers $$a_i=1-\Lambda^{-2i} \ \ \mbox{and} \ \ b_i=1+\Lambda^{-2i+1} $$ Then $$ \frac{b_i-a_i}{b_i-a_j} = \frac{\Lambda +1}{\Lambda + \Lambda^{2(i-j)}}$$ When $j<i$ we have $$ 0 \leq \frac{b_i-a_i}{b_i-a_j} \leq \frac{\Lambda+1}{\Lambda+\Lambda^2} \leq \frac{1}{\Lambda}$$ while for $j\geq i$ $$ 1\leq \frac{b_i-a_i}{b_i-a_j} \leq 1+\frac{1}{\Lambda}$$ Therefore $$n-i+1 \leq \frac{1}{M_i}\leq n-i+1 + \frac{n}{\Lambda} $$ Letting $\Lambda$ go to infinity we obtain: $$ \lim_{\Lambda\rightarrow \infty} \ \sum_i {M_i} = \sum_{k=1}^n \frac{1}{k}$$ thus showing that the stated bound is optimal.


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.