0
$\begingroup$

My textbook gives the following example:

The second-order cone is the norm cone for the Euclidean norm, i.e.,

$$\begin{align} C &= \{ (x, t) \in \mathbb{R}^{n + 1} \mid \left\lVert x \right\rVert_2 \le t \} \\ &= \left\{ \begin{bmatrix} x \\ t \end{bmatrix} \;\middle|\; \begin{bmatrix} x \\ t \end{bmatrix}^T \begin{bmatrix} I & 0 \\ 0 & -1 \end{bmatrix} \begin{bmatrix} x \\ t \end{bmatrix} \le 0, t \ge 0 \right\} \end{align}$$

The second-order cone is also known by several other names. It is called the quadratic cone, since it is defined by a quadratic inequality. It is also called the Lorentz cone or ice-cream cone.

I'm wondering how to multiply out $$\begin{bmatrix} x \\ t \end{bmatrix}^T \begin{bmatrix} I & 0 \\ 0 & -1 \end{bmatrix} \begin{bmatrix} x \\ t \end{bmatrix}$$

It seems to me that I need to use block matrix multiplication (assuming $I$ is the identity matrix), but even after reviewing examples (see this and Wikipedia), I keep getting the matrix $\begin{bmatrix} x^2 I - t^2 \end{bmatrix}$ as my final result, which doesn't make sense, since $x^2 I$ is an invalid matrix multiplication.

I would greatly appreciate it if people could please take the time to demonstrate this.

$\endgroup$
1
$\begingroup$

$\newcommand{\R}{\mathbb{R}}$Let $ \mathbf{0}\in\R^n$ be a column of all $0$'s, so $\mathbf{0}^T$ denotes a row of $n$ zeros. Note that by block matrix multiplication, we have $$\begin{bmatrix} I & \mathbf{0} \\ \mathbf{0}^T & -1\end{bmatrix}\begin{bmatrix} \mathbf{x} \\ t\end{bmatrix} = \begin{bmatrix}I\mathbf{x} + \mathbf{0} t\\ \mathbf{0}^T \mathbf{x} +(-1)t\end{bmatrix} = \begin{bmatrix}\mathbf{x} \\ -t\end{bmatrix}\in \R^{n+1}.$$ Also, $\begin{bmatrix}\mathbf{x} \\ t\end{bmatrix}^T = \begin{bmatrix}\mathbf{x}^T & t\end{bmatrix}\in\R^{1\times (n+1)}$ and so

$$\color{blue}{\begin{bmatrix}\mathbf{x} \\ t\end{bmatrix}^T \begin{bmatrix}\mathbf{x} \\ -t\end{bmatrix} = \begin{bmatrix}\mathbf{x}^T & t\end{bmatrix} \begin{bmatrix}\mathbf{x} \\ -t\end{bmatrix} = \mathbf{x}^T \mathbf{x} - t^2}.$$

(It looks like your error was maybe in the above step.)

Combining these facts tells us that $$\boxed{\begin{bmatrix}\mathbf{x} \\ t\end{bmatrix}^T \begin{bmatrix} I & \mathbf{0} \\ \mathbf{0}^T & -1\end{bmatrix}\begin{bmatrix} \mathbf{x} \\ t\end{bmatrix} = \mathbf{x}^T \mathbf{x} - t^2}.$$

Note that this means the inequality $$\begin{bmatrix}\mathbf{x} \\ t\end{bmatrix}^T \begin{bmatrix} I & \mathbf{0} \\ \mathbf{0}^T & -1\end{bmatrix}\begin{bmatrix} \mathbf{x} \\ t\end{bmatrix} \le 0$$ becomes $$\mathbf{x}^T \mathbf{x} - t^2\le 0,$$ which is equivalent to $\left\|\mathbf{x}\right\|_2 \le t$ (for $t\ge 0$), as was claimed.

$\endgroup$
  • $\begingroup$ Thanks for the answer. Can you please explain why we require that $\mathbf{0}^T$ denote a row of $n$ zeroes? $\endgroup$ – The Pointer Apr 5 at 23:31
  • $\begingroup$ That is the meaning of the lower-left "0" in your original post's matrix (becausethe overall matrix is $(n+1)\times (n+1)$. Also the other (top-right) 0 in that matrix refers to a column of $n$ zeros. It's a good idea to keep track of the dimensions of things in these block matrix problems. $\endgroup$ – Minus One-Twelfth Apr 5 at 23:51
  • $\begingroup$ Ahh, I didn’t realize that. So the author has already partitioned the matrix correctly to allow for easy computation? $\endgroup$ – The Pointer Apr 5 at 23:52
  • $\begingroup$ Correct! $\ddot{\smile}$ $\endgroup$ – Minus One-Twelfth Apr 5 at 23:53

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.