Another way to do this: we want to minimize the distance from (0, 0, 0) to (x, y, z) subject to the condition that [itex]xyz^2= 2[/itex]. You say you know that "the shortest line between the origin to the surface will also be normal to the surface." Okay, the normal to that surface at any point will be in the direction of the vector [itex]\nabla (xyz^2)= yz^2\vec{i}+ xz^2\vec{j}+ 2xyz\vec{k}[/itex]. And any line through the origin to point (x, y, z) is in the direction [itex]x\vec{i}+ y\vec{j}+ z\vec{k}[/itex].
In order that the line through the origin be normal to the surface, those two vectors must be parallel which means one is a multiple of the other. Calling that multiplier "[itex]\lambda[/itex]". Then we must have
[itex]yz^2\vec{i}+ xz^2\vec{j}+ 2xyz\vec{k}= \lambda(x\vec{i}+ y\vec{j}+ z\vec{k})[/itex]
(this is equivalent to the "Lagrange multiplier" method).
So [itex]yz^2= \lambda x[/itex], [itex]xz^2= y[/itex], and [itex]2xyz= \lambda z[/itex] as well as the original equation of the surface, [itex]xyz^2= 2[/itex]. That gives four equations to solve for x, y, z, and [itex]\lambda[/itex].
Since a value for [itex]\lambda[/itex] is not part of the solution, I find it typically best to eliminate [itex]\lambda[/itex] first by dividing one equation by another. Dividing the first equation by the second,
[tex]\frac{yz^2}{xz^2}= \frac{\lambda x}{\lambda y}[/tex]
[tex]\frac{y}{x}= \frac{x}{y}[/tex]
so [tex]y^2= x^2[/tex] so y= x or y= -x.
Similarly, dividing the second equation by the third
[tex]\frac{xz^2}{2xyz}= \frac{\lambda y}{\lambda z}[/tex]
[tex]\frac{z}{2y}= \frac{y}{z}[/tex]
[itex]z^2=2y^2[/itex] so [itex]z= \sqrt{2}y[/itex] or [itex]z= -\sqrt{2}y[/itex].
If [itex]x= y[/itex] and [itex]z= \sqrt{2}y= \sqrt{2}x[/itex] then [itex]xyz^2= x(x)(2x^2)= 2x^4= 2[/itex] so x= y= z= 1 or x= y= z= -1.
Do the other possibilities, [itex]y= -x[/itex] and/or [itex]z= -\sqrt{2}y[/itex] in the same way.