The problem asks to prove the arithmetic-geometric mean inequality $(a_{1}a_{2}...a_{n})^{1/n} \leq \frac{1}{n}(a_{1}+a_{2}+...+a_{n})$ for $a_{i}>0$ by minimizing the function $f(\vec{x})=\frac{1}{n}(x_{1}+x_{2}+...+x_{n})$ on the surface $g(\vec{x})=x_{1}x_{2}...x_{n}-1=0$.
I found that the minimum occurs when all $x_{i}=1$ so $f(1,...,1)=1$. So now we have $\frac{1}{n}(x_{1}+...+x_{n}) \geq1$ if $x_{1}x_{2}...x_{n}=1$.
I think my work up to this point is correct., but I'm really not sure how to use this fact to prove the G-A inequality. I thought about putting $x_{i}$ in terms of $a$ somehow so their product is always $1$ but I'm just not getting it.
In general given arbitrary positive numbers, $y_1$, $y_2, \ldots, y_n$.
Define $z_i = \frac{y_i}{\left(\prod_{k=1}^n y_k\right)^\frac1n}$
then we have $\prod_{i=1}^n z_i =1$, by your conclusion,
$$\frac{1}{n}\sum_{i=1}^n z_i \ge 1$$
which is $$\frac{1}{n}\sum_{i=1}^n y_i \ge \left(\prod_{i=1}^ny_i\right)^\frac1n$$