I am looking for a nice(er) proof of the following identity: $$\sum_{k=1}^{n-1} (-1)^k \cos^{2j}\frac{\pi k}{2n} = -\frac{1}{2}, ~ 0<j<n,~j,n\in \mathbb{Z}^+.$$ I have been able to prove it using the following strategy. Express the cosine in terms of the complex exponential, obtaining the summand $(-1)^k\left(1/2+e^{i\pi k/n}/4+e^{-i\pi k/n}/4\right)^j$. Use the trinomial expansion to expand the bracket, then sum over $k$ which is now a geometric series. In particular, I find $$-\sum_{p+q\leq j} {j\choose p~q} \frac{1}{2^{j-p-q}}\frac{1}{4^p}\frac{1}{4^q} \frac{(-1)^{n+p-q}+e^{i(p-q)\pi/n}}{1+e^{i(p-q)\pi/n}}=-\sum_{p+q\leq j} {j\choose p~q} \frac{1}{2^{j-p-q}}\frac{1}{4^p}\frac{1}{4^q}\times\begin{cases}1, ~ n+p-q=\text{even}\\0, ~ n+p-q=\text{odd}\end{cases},$$ where the odd case follows as the result would be imaginary. The trinomial then reconstitutes to $$-(1/2+1/4+1/4)^j/2 \pm (1/2-1/4-1/4)^j/2=-1/2,$$ where the choice of sign depends on the parity of $n$.
Is there some deeper interpretation of this identity, and/or a slicker way to prove it? There is some clear overlap with Sum of Fourth Powers of cosine series has closed form solution., epecially the answer given there by Claude Leibovici.