Monotonically decreasing function

calculusmonotone-functionsreal-analysis

My goal is to prove that the function $f(x) = (n \cot{(x)} – 2 \cot{(2x/n)}) \csc{(2x/n)}$ is monotonically decreasing on interval $0 < x \leq \pi/2$ for any $n > 2$. I'm working with two different approaches.

For the first, consider the expansion of the function given by Wolfram Alpha
\begin{equation}
f(x) = \frac{1}{6} (4 – n^2) + \frac{1}{90} \left(-n^2 + \frac{56}{n^2} – 10\right) x^2 – \frac{n^6 + 7n^4 + 49n^2 – 372}{945n^4} x^4 – \frac{3n^8 + 20n^6 + 98n^4 + 620n^2 – 6096}{28350n^6} x^6 + O(x^8),
\end{equation}

valid until $\pi$. All terms considered are something negative for $n > 2$. If there is case for all terms of the expansion, the function is monotonically decreasing. Of course, with a finite number of terms, we didn't prove anything. However, is
is it possible to somehow generalize the conclusion? It seems to very difficult to obtain an analytical expression for the expansion of $f(x)$.

The second approach is the usual analysis of the nonpositivity of the derivative. Compute derivative gives
\begin{equation}
\frac{\csc{\left(2x/n\right)} \left(-2n \cot{(x)} \cot{\left(\frac{2x}{n}\right)} + 4 \cot^2{\left(2x/n\right)} – n^2 \csc^2{(x)} + 4 \csc^2{\left(2x/n\right)}\right)}{n}
\end{equation}

Since they are positive, we can ignore the factor $\csc{(2 x/n)}$ and the denominator. With some manipulations
\begin{equation}
\csc^2{(x)} \csc^2{\left(2x/n\right)} \left(4(1 + \cos^2{\left(2x/n\right)}) \sin^2{(x)} – n^2 \sin^2{\left(2x/n\right)} – \frac{1}{2}n \sin{(2x)} \sin{\left(4x/n\right)}\right)
\end{equation}

Ignore positive factors $\csc^2{(x)}$ and $\csc^2{(2x/n)}$ gives a function in terms of sines and cosines. I don't know how to simplify it. I try to use upper and lower bound to get a simple expression but truncated Taylor series bounds $\sin{(x)} \geq x – x^3/6$ and $\cos{(x)} \leq 1 – x^2/2 + x^4/24$ but the expression becomes even more complicated (unfortunately, $\cos{(x)} \leq 1$ results in a non-monotonic function). There's a simpler approximation for sine and cosine?

Best Answer

You have a more than decent approximation of the function if you use the $[2n,2n]$ Padé approximant $P_n$ of it.

For example $$P_1=\frac 16\,\,\frac {\left(4-n^2\right)+\frac{\left(3 n^6-98 n^4-98 n^2+1768\right) }{105 n^2 \left(n^2+14\right)}x^2 } {1-\frac{2 \left(n^4+11 n^2+93\right) }{21 n^2 \left(n^2+14\right)}x^2 }$$

To give an idea, consider the infinite norm $$\Phi_n=\int_0^{\frac \pi 2} \Big(f(x)-P_1\Big)^2\,dx$$

A few values $$\left( \begin{array}{cc} n & \Phi_n \\ 3 & 5.6139\times 10^{-6} \\ 4 & 3.3417\times 10^{-7} \\ 5 & 6.6111\times 10^{-7} \\ 6 & 3.8840\times 10^{-6} \\ 7 & 9.1023\times 10^{-6} \\ 8 & 1.6175\times 10^{-5} \\ \end{array} \right)$$

This should help since the numerator of the derivative of $P_1$ with respect to $x$ is $$-(n-2) n^2 (n+2) \left(n^2+14\right)^3 x$$