Goal: prove that at each point point in the n-dimensional Euclidean space, there passes precisely n confocal quadrics; they are mutually orthogonal.

Existence Proof:

We will prove this for n = 3; the proof can be generalized to any n.

Given a point (x_0, y_0, z_0) \in \mathbb{R}^3. The family of confocal quadrics can be given by:

\frac{x^2}{a - \lambda} + \frac{y^2}{b - \lambda} + \frac{z^2}{c - \lambda} = 1, for fixed a, b, c \in \mathbb{R}.

To show that 3 distinct mutually confocal quadrics pass through (x_0, y_0, z_0), it suffices to show that \lambda has 3 distinct roots in the equation:

\frac{x_0^2}{a - \lambda} + \frac{y_0^2}{b - \lambda} + \frac{z_0^2}{c - \lambda} = 1

Without loss of generality, suppose a \leq b \leq c. How do we show that there are 3 distinct roots? We will not be computing any of the solutions using algebra. Instead, we will use the intermediate value theorem to show that the function f(\lambda) =   \frac{x_0^2}{a - \lambda} + \frac{y_0^2}{b - \lambda} + \frac{z_0^2}{c - \lambda} crosses y = 1 exactly 3 times.

Key observation: f(\lambda) has 3 vertical asymptotes. We need to analyze the behaviour of the function to the left and to the right of each asymptote. We will then obtain:

Done!

Orthogonality: The n mutually confocal quadrics intersecting at a point are orthogonal to each other.

Again, we prove this for n = 3; generalization to any n \in \mathbb{N} can be made.

Suppose

\frac{x^2}{a - \lambda_1} + \frac{y^2]{b - \lambda_1} + \frac{z^2}{c - \lambda_1} = 1 (*)

and

\frac{x^2}{a - \lambda_2} + \frac{y^2]{b - \lambda_2} + \frac{z^2}{c - \lambda_2} = 1 (**)

are 2 quadrics passing through the point (x_0, y_0, z_0).

Note that (*) is a level curve of f(x, y, z) =  \frac{x^2}{a - \lambda_1} + \frac{y^2]{b - \lambda_1} + \frac{z^2}{c - \lambda_1} and (**) is a level curve of g(x, y, z) = \frac{x^2}{a - \lambda_2} + \frac{y^2]{b - \lambda_2} + \frac{z^2}{c - \lambda_2}.

It follows that \nabla f(x, y, z)|_{(x_0, y_0, z_0)} is normal to the surface defined by (*) and \nabla g(x, y, z)|_{(x_0, y_0, z_0)} is normal to the surface defined by (**) at (x_0, y_0, z_0).

Hence, to show that the two surfaces are orthogonal to each other, it suffices to show that the two corresponding gradient vectors are orthogonal.

\nabla f(x, y, z)|_{(x_0, y_0, z_0)} = (\frac{2x_0}{a - \lambda_1}, \frac{2y_0}{b - \lambda_1}, \frac{2z_0}{c - \lambda_1}) and

\nabla g(x, y, z)|_{(x_0, y_0, z_0)} = (\frac{2x_0}{a - \lambda_2}, \frac{2y_0}{b - \lambda_2}, \frac{2z_0}{c - \lambda_2}).

Now, how do we show that \nabla f(x, y, z)|_{(x_0, y_0, z_0)} \cdot \nabla g(x, y, z)|_{(x_0, y_0, z_0)} = 0?

Key: f(x_0, y_0, z_0) - g(x_0, y_0, z_0) = 0

\Leftrightarrow (\frac{x_0^2}{a - \lambda_1} + \frac{y_0^2]{b - \lambda_1} + \frac{z_0^2}{c - \lambda_1}) - (\frac{x_0^2}{a - \lambda_2} + \frac{y_0^2]{b - \lambda_2} + \frac{z_0^2}{c - \lambda_2}) = 0

\Leftrightarrow (\lambda_1 - \lambda_2)(\frac{x_0^2}{(a - \lambda_1)(a - \lambda_2)} + \frac{y_0^2}{(b - \lambda_1)(b - \lambda_2)} + \frac{z_0^2}{(c - \lambda_1)(c - \lambda_2)}) = 0 (***).

By (***),

\nabla f(x, y, z)|_{(x_0, y_0, z_0)} \cdot \nabla g(x, y, z)|_{(x_0, y_0, z_0)} = 2(\frac{x_0^2}{(a - \lambda_1)(a - \lambda_2)} + \frac{y_0^2}{(b - \lambda_1)(b - \lambda_2)} + \frac{z_0^2}{(c - \lambda_1)(c - \lambda_2)}) = 0.

This concludes the proof.

Something I’m currently thinking about: how do we construct these family of quadrics in the higher dimensions? For n = 2, given 2 foci, the ellipse and the hyperbola are formed by considering points for which the sum of the distances from the foci or the difference of such distances is a fixed value. What about for n = 3?

Leave a Reply

Is this your new site? Log in to activate admin features and dismiss this message
Log In