It's vec(grad)p(x,y) = langle-(4x)/(sqrt(24-4x^2-y^2)) , -(y)/(sqrt(24-4x^2-y^2)) rangle
The gradient vec(grad)p(x,y) of the function p(x,y) is defined as the vector whose n-th component is the partial derivative of p with respect to the n-th variable, or:
vec(grad)p(x,y) = langle (del p)/(del x), (del p)/(del y) rangle
Computing the partial derivatives using the chain rule:
(del p)/(del x) = (del)/(del x)[sqrt(24-4x^2-y^2)]
= 1/(2sqrt(24-4x^2-y^2)) (del)/(del x)[24-4x^2-y^2]
= -(4x)/(sqrt(24-4x^2-y^2))
(del p)/(del y) = (del)/(del y)[sqrt(24-4x^2-y^2)]
= 1/(2sqrt(24-4x^2-y^2)) (del)/(del y)[24-4x^2-y^2]
= -(y)/(sqrt(24-4x^2-y^2))
And so,
vec(grad)p(x,y) = langle (del p)/(del x), (del p)/(del y) rangle
= langle -(4x)/(sqrt(24-4x^2-y^2)) , -(y)/(sqrt(24-4x^2-y^2)) rangle