# Contraction maps with $3\times 3$ matrices, choice for $x$?

$f:\Bbb R^2\to \Bbb R^2$ given by $$f(x)=\begin{bmatrix}\frac12&0\\0&\frac13\end{bmatrix}\begin{bmatrix}x_1\\x_2\end{bmatrix}$$ We can determine if this is a contraction map by showing that $\|Ax\|\leq k \lt 1$ when $\|x\|=1$

Therefore we can take $x=(\cos t,\sin t)$

and hence $$Ax=\begin{pmatrix}\frac12 \cos t\\\frac13\sin t\end{pmatrix}$$ $$\|Ax\|^2=\frac14\cos^2t+\frac19\sin^2 t\leq \frac14$$ And hence $\|Ax\|\leq\frac12$, and thus this is a contraction.

What about in the case where we have a $3\times 3$ matrix? $$f(x)=\begin{bmatrix}\frac12&0&\frac13\\0&\frac14&0\\-\frac12&0&\frac13\end{bmatrix}\begin{bmatrix}x_1\\x_2\\x_3\end{bmatrix}$$ $f:\Bbb R^3\to \Bbb R^3$, can we still just test for $\|x\|=1$ and what do we choose for $x$? We need three terms I guess?

For your $3 \times 3$ problem: note that this is really just the $2 \times 2$ problem in disguise. In particular, we note that $$\left\|\pmatrix{ \frac12&0&\frac13\\ 0&\frac14&0\\ -\frac12&0&\frac13 } \pmatrix{x_1\\x_2\\x_3}\right\|^2 = \left\|\pmatrix{ \frac12&\frac13\\ -\frac12&\frac13 }\pmatrix{x_1\\x_3}\right\|^2 + (x_2/4)^2$$ From there, the technique you used last time applies.

I suspect that if they give you a $3 \times 3$ question on the exam, you can break it down like this.

Complete solution:

We see that $$\left\|\pmatrix{ \frac12&\frac13\\ -\frac12&\frac13 }\pmatrix{x_1\\x_3}\right\|^2 = \left\| \pmatrix{(1/2) x_1 + (1/3)x_3\\ (-1/2)x_1 + (1/3)x_3} \right\|^2 =\\ [(1/2) x_1 + (1/3)x_3]^2 + [(-1/2)x_1 + (1/3)x_3]^2 =\\ [x_1^2/4 + x_1x_3/3 + x_3^2/9] + [x_1^2/4 - x_1x_3/3 + x_3^2/9] =\\ (1/2)x_1^2 + (2/9)x_3^2 \leq\\ (1/2)x_1^2 + (1/2)x_3^2=\\ (1/2)\|(x_1,x_3)\|^2$$ So, we have $$\left\|\pmatrix{ \frac12&0&\frac13\\ 0&\frac14&0\\ -\frac12&0&\frac13 } \pmatrix{x_1\\x_2\\x_3}\right\|^2 = \left\|\pmatrix{ \frac12&\frac13\\ -\frac12&\frac13 }\pmatrix{x_1\\x_3}\right\|^2 + (x_2/4)^2 \leq\\ (1/2)\|(x_1,x_3)\|^2 + (1/4)x_2^2 \leq\\ (1/2)\|(x_1,x_3)\|^2 + (1/2)x_2^2 =\\ (1/2)(x_1^2 + x_2^2 + x_3^2) =\\ (1/2)\|(x_1,x_2,x_3)\|^2$$ We conclude that $f$ is a contraction with $k = \frac{1}{\sqrt{2}}$.

• Oh wow, did you see that immediately, or you just realised it? Also what would be my smart choice for $x=(x_1,x_2,x_3)$ here? – Calculus Jun 17 '15 at 5:37
• What you suspect is probably true, since the only two past exams I can view can be – Calculus Jun 17 '15 at 5:39
• I noticed that besides the diagonal entry, all the entries in the second row and second column were $0$. – Omnomnomnom Jun 17 '15 at 5:39
• I would set $x_1 = a\cos t, x_3= a\sin t$ with $a \in (0,1)$. Then you just need to remember that if $\|x\| = 1$, then $x_2^2 = 1-\|(x_1,x_3)\|^2$. – Omnomnomnom Jun 17 '15 at 5:41
• The neat thing is that it is enough to show that the individual maps $$x \mapsto \pmatrix{ \frac12&\frac13\\ -\frac12&\frac13 }x\\ x \mapsto x/4$$ are contractions on $\Bbb R^2$ and $\Bbb R$, respectively. – Omnomnomnom Jun 17 '15 at 5:44

Yes, you can still use $\|x\|=1$. Yes, you will need 3 terms.

I would start by naively checking the norm of cos x, sin x, cos x

• Why is the third term $\cos x$? I imagined the choice in the $\Bbb R^2$ case was because $x=\cos t + i\sin t=(\cos t,\sin t)$ and thus we were looking at $\Bbb C$, so I guessed that we would want spherical coordinates or something? – Calculus Jun 17 '15 at 5:13
• @Calculus You could definitely use spherical coordinates, that is $$(\sin s \cos t, \sin s \sin t, \cos s)$$ – Omnomnomnom Jun 17 '15 at 5:21
• @Omnomnomnom Is this a reliable method in general though? It seems the hand-computations will be very tedious(especially in an exam!) side-note: my class taught us only how to do it with $2\times 2$ matrices, but all past exams(with no solution keys) have $3\times 3$ matrices!! – Calculus Jun 17 '15 at 5:24
• @Calculus do you know about eigenvalues? – Omnomnomnom Jun 17 '15 at 5:25
• @Omnomnomnom Yes, we can compute these by finding the characteristic equation, or if we are lucky(upper or lower triangular I believe) we can read them off(maybe something to do with being orthogonal here too?) – Calculus Jun 17 '15 at 5:25

More generally, suppose you have a $3\times 3$ matrix $A$ whose column vectors are orthogonal vectors $\mathbf u$, $\mathbf v$, and $\mathbf w$. Then I claim that $\|A\mathbf x\| \le k\|\mathbf x\|$ for $k=\max(\|\mathbf u\|,\|\mathbf v\|,\|\mathbf w\|)$. For note that $$A\mathbf x = x_1\mathbf u + x_2\mathbf v + x_3\mathbf w$$ and so \begin{align*} \|A\mathbf x\|^2 &= \|x_1\mathbf u + x_2\mathbf v + x_3\mathbf w\|^2 \\ &= \big(x_1\mathbf u + x_2\mathbf v + x_3\mathbf w\big)\cdot\big(x_1\mathbf u + x_2\mathbf v + x_3\mathbf w\big) \\ &= \|\mathbf u\|^2x_1^2 + \|\mathbf v\|^2x_2^2 + \|\mathbf w\|^2x_3^2 \\ &\le k^2(x_1^2+x_2^2+x_3^2) = k^2\|\mathbf x\|^2. \end{align*}