Given any $n$th degree polynomial $P(x)$ with positive coefficients and positive numbers $a, b\in\mathbb{R}$, does $\sqrt[n]{P(a+b)} \leq \sqrt[n]{P(a)} + \sqrt[n]{P(b)}$ hold?
I've worked it out for $n=2$:
$$P(x) = Ax^2 + Bx + C$$
$$\sqrt{Aa^2 + Ba + C} + \sqrt{Ab^2 + Bb + C} = \sqrt{(\sqrt{Aa^2 + Ba + C} + \sqrt{Ab^2 + Bb + C})^2} $$
$$= \sqrt{(Aa^2 + Ba + C) + 2\sqrt{(Aa^2 + Ba + C)(Ab^2 + Bb + C)} + (Ab^2 + Bb + C)}$$
$$ \geq \sqrt{Aa^2 + Ba + C + 2\sqrt{(Aa^2)(Ab^2)} + Ab^2 + Bb + C}$$
$$ \geq \sqrt{Aa^2 + Ba + C + 2Aab + Ab^2 + Bb}$$
$$ = \sqrt{A(a^2 + b^2 + 2ab) + B(a+b) + C} = \sqrt{A(a+b)^2 + B(a+b) + C}$$
But I couldn't figure out the generalization from there.
I also pondered whether Jensen's inequality would be of help, but I don't think $\sqrt[n]{P(x))}$ is concave.
I have used various higher degree polynomials to plot $\sqrt[n]{P(x))}$ and the inequality always seems to hold.