Find the $3 \times 3$ matrix $ A$, associated with the linear transformation that projects vectors in $\mathbb{R^3}$ (orthogonally) onto the plane $x+y+z=0$.

I was given this question just as a review question for my class. I took linear algebra over 2 years ago, so my memory is really fuzzy.

I was given the hints to find the matrix $A$ by thinking of it as a composition of a rotation, simpler projection, then another rotation.

I know the normal vector of the plane is $<1,1,1>$. I think this may be a useful fact. Also after googling a bit I found that the matrix enter image description here. Perhaps, I could work with this, although I honestly don't really know exactly how to proceed. I know there are many projections onto this plane, but I don't know exactly how to find an orthogonal one. Any help would be much appreciated.


A projection matrix onto a plane in 3-D Euclidean space always has the form $$\mathbf P = \mathbf I - \frac{\mathbf n \mathbf n^T}{\mathbf n^T \mathbf n},$$

where $\mathbf n$ is the normal vector on the plane. Let's see why this is true. What do we expect if we apply $\mathbf P$ to $\mathbf n$? We should get zero right? Well, as you can verify we do, since $$\mathbf P \mathbf n = \mathbf n - \frac{\mathbf n \mathbf n^T\mathbf n}{\mathbf n^T \mathbf n} = \mathbf n - \mathbf n = 0.$$ What happens to any other vector $\mathbf a$? We get $$\mathbf P \mathbf a = \mathbf a - \mathbf n\frac{ \mathbf n^T\mathbf a}{\mathbf n^T \mathbf n},$$ which is exactly the projection formula: subtract from $\mathbf a$ the piece of $\mathbf n$ that is as long as the projection of $\mathbf a$ onto $\mathbf n$. If you draw this in 2-D, it should become clear quite quickly.

Here is an illustration: Projection in 2-D To project $\mathbf a$ on a "plane" (in 2-D, a line $\mathbf p$), subtract from $\mathbf a$ the blue vector, which is just like the normal vector $\mathbf n$ but scaled to the length of $\mathbf a$ projected onto $\mathbf n$.

Now, with this formula, since you already have your $\mathbf n$, the rest should be easy...

  • $\begingroup$ Thanks for the answer. In the formula is n a column vector, and I the identity matrix? I'm not exactly sure how to calculate nnT/ nTn. I don't really know how to divide matrices. Thanks. $\endgroup$ – kemb Jan 19 '18 at 0:31
  • $\begingroup$ @kemb: Yes, $\mathbf n$ is a column vector and $\mathbf I$ is an identity matrix. Since $\mathbf n$ is a column vector, $\mathbf n^T \mathbf n$ is a scalar, in fact, equal to the squared Euclidean norm. So the division is, of course, just a scalar division. $\endgroup$ – Florian Jan 19 '18 at 15:23
  • $\begingroup$ @kemb: did you suceed with answering the question? Is anything still unclear? If the answer was helpful to you, consider accepting it as correct. $\endgroup$ – Florian Jan 22 '18 at 13:49

All you have to do is to find where the transformation sends the basic unit vectors $e_1=\pmatrix{1\\0\\0}$, $e_2$ and $e_3$ to, since these will be the columns of your transformation matrix.

A normal vector to the plane $x+y+z=0$ is $n=\pmatrix{1\\1\\1}$ so $e_1$ will be sent to $e_1+\alpha n=\pmatrix{1+\alpha\\\alpha\\\alpha}$ for some $\alpha$. What then should $\alpha$ be?

  • $\begingroup$ Hello, Thanks. I'm not entirely sure what you are getting at. What does a represent and how do we find a? $\endgroup$ – kemb Jan 18 '18 at 5:23
  • $\begingroup$ If I understand you correctly, is a=0? $\endgroup$ – kemb Jan 18 '18 at 5:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.