0

Suppose an ellipsoid given by $\{{(x,y,z)| {x^{2}+\frac{y^{2}}{4}+\frac{z^{2}}{9}}}=1\}$, find the area of the orthogonal projection of the ellipsoid on the plane ${2x+4y-5z=10}$.

What is the right direction to solve this question? Thanks in advance!

  • 2
    Presumably you meant $x^2+y^2/4+z^2/9\le1$ since otherwise your set-building expression is meaningless. Think about what the outline of the projected image is. – amd Jan 14 '19 at 21:10
  • What's the logic behind closing this kind of question? It is interesting. – orangeskid Jul 12 '22 at 12:26

2 Answers2

2

Without calculus: The outline of the image of a quadric $Q$ under the central projection with matrix $P$ is a conic with a simple formula expressed via dual conics/quadrics: $C^*=PQ^*P^T$. What this formula says is that the outline of the image of $Q$ is the conic that is tangent to the images of the tangent planes of $Q$ that pass through the center of projection.

If we apply a rotation $R$ that aligns the projection direction with the $z$-axis, our orthogonal projection will then have the matrix $$P = \begin{bmatrix}1&0&0&0\\0&1&0&0\\0&0&0&1\end{bmatrix} R,$$ i.e., the projection consists of rotating and then dropping the $z$-coordinate. The first two rows of $P$ are therefore the first two rows of $R$, for which we can take any two orthogonal unit vectors perpendicular to the image plane’s normal. $(4,-2,0)^T$ is obviously one such vector, and we can obtain another via a cross product: $(2,4,-5)^T\times(4,-2,0)^T=(-10,-20,-20)^T$. Normalizing these two vectors gives us $$P=\begin{bmatrix}{2\over\sqrt5} & -{1\over\sqrt5} & 0 & 0 \\ -\frac13 & -\frac23 & -\frac23 & 0 \\ 0&0&0&1 \end{bmatrix}.$$ The matrix $Q = \operatorname{diag}(1,1/4,1/9,-1)$, so using $Q^{-1}$ for the dual conic $Q^*$ is convenient. Multiplying this all out produces $$C^* = \begin{bmatrix}\frac85 & {2\over\sqrt5} & 0 \\ {2\over\sqrt5} & \frac{53}9 &0 \\ 0&0&-1 \end{bmatrix}.$$ The inverse of this matrix represents an ellipse centered at the origin.

There’s no need to invert this matrix to obtain the equation of the ellipse, though, since the ellipse’s area can be computed directly from the dual matrix. The area of an ellipse is $\pi$ times the product of its semiaxis lengths, and the latter are the reciprocal square roots of the eigenvalues of the quadratic part of $C$. Therefore, the area of this ellipse is equal to $${\pi\over\sqrt{-\det C}} = \pi \sqrt{-\det C^*} = \frac23\sqrt{97\over5}\pi.$$

With calculus: One possible approach is to find a parameterization of the outline of the projection. The generator of the outline of a quadric surface—the locus of intersection of rays from the projection center tangent to the surface—it itself a conic that lies in the polar plane of the projection center. For this parallel projection the center is the point at infinity $\mathbf V=(2,4,-5,0)$, so the generator’s plane is $Q\mathbf V = (2,1,-5/9,0)^T$, i.e., $2x+y-\frac59z=0$. Since the ellipsoid is axis-aligned, a relatively easy way to find a parameterization for the intersection curve is to transform the surface to a unit sphere. This transformation changes the polar plane to $(2,2,-5/3,0)^T$. Proceeding as above, we find a pair of unit vectors orthogonal to this plane’s normal, leading to the circle $$\cos(t)\left(\frac1{\sqrt2},-\frac1{\sqrt2},0\right)^T+\sin(t)\left(-{5\over\sqrt{194}},-{5\over\sqrt{194}}-6\sqrt{2\over97}\right)^T.$$ Mapping this circle back to the original coordinate system produces $$\cos(t)\left(\frac1{\sqrt2},-\sqrt2,0\right)^T+\sin(t)\left(-{5\over\sqrt{194}},-5\sqrt{2\over97},-18\sqrt{2\over97}\right)^T.$$

Since we’re working with a parallel projection, we can use any plane parallel to the image plane, so we project this onto $2x+4y-5z=0$. This is equivalent to projecting onto the original image plane and then translating the center of the resulting ellipse to the origin. The projection onto this plane is easily computed via the orthogonal rejection from the plane’s normal, resulting finally in a parameterization $\mathbf r(t)$ for the outline of the ellipsoid’s projection. Its area is then equal to $\frac12\int_0^{2\pi}\|\mathbf r\times d\mathbf r\|$. You should find that the integrand simplifies nicely.

On the other hand, once we had the two unit vectors that generated the great circle of the unit sphere above, we could again proceed without calculus. Scaling and projecting these vectors produces the halves of a pair of conjugate diameters of the outline. A theorem of Appolonios states that the area of the triangle formed by them is equal to $\frac12ab$, and so the area of the ellipse is equal to $2\pi$ times this area. Calling these two vectors $\mathbf u$ and $\mathbf v$, the area of ellipsoid’s projection is therefore equal to $\pi\|\mathbf u\times\mathbf v\|$.

amd
  • 55,457
  • How can we do it with calculus? – alexander_yz Jan 15 '19 at 22:09
  • @alexander_yz Read the penultimate paragraph. It’s also possible to compute this by integrating $\mathbf n\cdot dS$, where $\mathbf n$ is a unit normal to the image plane, over an appropriate piece of the ellipsoid, but getting the bounds right is not easy. – amd Jan 16 '19 at 00:39
  • the bounds are my main problem – alexander_yz Jan 16 '19 at 11:33
  • @alexander_yz I would look for a change of coordinates such that you’re integrating over a hemisphere. It will involve much of the same scaling and rotation that’s in this answer. – amd Jan 16 '19 at 17:47
  • Thanks a lot! :) – alexander_yz Jan 20 '19 at 08:49
0

Given the ellipsoid

$$ C\to p'Mp=1,\ \ \ p = (x,y,z), \ \ \ M = \mbox{diag}\left({1,\frac 14, \frac 19}\right) $$

and the plane

$$ \Pi\to(p-p_0)\cdot\vec v,\ \ \ \vec v = (2,4,-5) $$

the line

$$ L\to p=p_1+\lambda \vec v $$

is by construction, orthogonal to $\Pi$. Now making

$$ (p_1+\lambda\vec v)'M(p_1+\lambda\vec v)=1 $$

and solving for $\lambda$ we have

$$ \lambda = \frac{-2p_1'M\vec v\pm 2\sqrt{(p_1'M\vec v)^2-(p_1'M p_1-1)(\vec v'M\vec v)}}{2\vec v'M\vec v} $$

but if $L$ is tangent to $C$ then

$$ (p_1'M\vec v)^2-(p_1'M p_1-1)(\vec v'M\vec v)=0 $$

now assuming that $p_1'$ is the generic tangency point, the sought projection is given by the solution

$$ (p_1'M\vec v)^2-(p_1'M p_1-1)(\vec v'M\vec v)=0\\ (p_1-p_0)'\cdot \vec v = 0 $$

for $p_1$

which eliminating $z$ gives

$$ 7028 x^2 + 1088 x y + 2837 y^2 = 9700 $$

this is the trace projection on the $z=0$ plane. Eliminating the $x y$ term with a convenient rotation we get

$$ (1973 + \sqrt{749929}) X^2 + (1973-\sqrt{749929}) Y^2 = 3880 $$

with main axes

$$ r_X=2 \sqrt{\frac{970}{1973+\sqrt{749929}}}\\ r_Y=2 \sqrt{\frac{970}{1973-\sqrt{749929}}} $$

so the projected area is given by

$$ S_p = \pi r_X r_Y = \frac{2 \sqrt{97} \pi }{9} $$

but

$$ S_p = S \left(\frac{\vec v}{||\vec v||}\cdot \hat e_z\right) $$

hence

$$ S = \frac{2 \sqrt{97} \pi }{45} $$

Attached the problem setup

enter image description here

In lightblue $(p_1'M\vec v)^2-(p_1'M p_1-1)(\vec v'M\vec v)=0$ in light red $(p_1-p_0)'\cdot \vec v = 0$ and in light yellow $7028 x^2 + 1088 x y + 2837 y^2 = 9700$

Cesareo
  • 36,905
  • What about the area? Aren't we supposed to make a surface integral? – alexander_yz Jan 14 '19 at 22:27
  • @alexander_yz After the contour, you can rotate it into the $x, y$ plane and then integrate it using polar coordinates. – Cesareo Jan 14 '19 at 22:36
  • @alexander_yz Attached a note about the area integration. – Cesareo Jan 15 '19 at 03:46
  • @alexander_yz Please. See the modifications to obtain the numerical area value. – Cesareo Jan 17 '19 at 15:32
  • I'm very thankful sir this helped me a lot! – alexander_yz Jan 19 '19 at 11:57
  • You can simplify this a bit by taking advantage of the relationship between the semiaxis lengths of the ellipse and eigenvalues of the quadratic form to compute the area directly from the translated ellipse equation without rotating it into standard position first. – amd Jan 20 '19 at 18:35
  • @amd Yes. I agree but I prefer to show explicitly some geometrical aspects that can be used as well. Mainly in a basic approach. – Cesareo Jan 20 '19 at 18:44