The Hessian matrix is
At each nonzero stationary point, its diagonal entries are and the off-diagonal entries are . Its eigenvalues are and , both positive. Thus the Hessian matrix is positive definite, and the second-derivative test proves that
Their common value is . In fact they are global minima: with , the arithmetic-geometric mean inequality gives , so
The last one-variable polynomial attains its minimum at , and equality in the first bound requires , precisely the four points above.
A stationary point must satisfy both factored equations from part (a). If either coordinate is zero, the other equation forces the other coordinate to vanish. This gives .
Otherwise and . With and , these become and . Dividing gives , so . Then forces . Hence all five stationary points are
The gradient is horizontal when and vertical when . Factoring those partial derivatives gives the two loci:
Thus each locus consists of an axis and two parabolas. Their common points are the five stationary points found below. At those points the gradient is zero and has no direction, so they are included in both algebraic zero-component conditions but excluded if “parallel” is required to mean a nonzero directed vector.
Figure 1.
Loci of horizontal and vertical gradients
. Blue curves have , orange curves have , and black points mark their intersections. At the black points the gradient itself vanishes. Details.
A necessary compatibility condition is , because a cross product with is perpendicular to it. If this condition fails, there is no solution.
If it holds, take the equation and use the vector triple product identity:
Thus the perpendicular orthogonal projection is fixed and the parallel component is free:
Conversely, under the compatibility condition, so every displayed vector solves the equation. The solution set is a straight line, including the line through the origin when .
The straight line consists of , with . Since the endpoints are distinct, the direction vector can be normalized. Choose
Putting gives the required parametric equation of a straight line . Other choices of on the same straight line, or the opposite direction for , give equivalent descriptions.
The scalar is the signed scalar projection along the unit vector . Multiplying by that unit vector makes the orthogonal projection of onto its span.
For the other expression, the vector triple product identity gives
This is the orthogonal projection onto a plane perpendicular to through the origin. Its dot product with is zero, while the difference from is parallel to .
From the factorisation in part (a), or . Substitution into the parabola equation gives respectively or . Thus the two intersections of the parabola and the unit circle are
Substitution for reduces the two simultaneous equations to
The remaining quadratic factor is positive for all real , since . Thus there are exactly .

There are unlisted articles, also show them or only show them.