Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Let $A$=\begin{pmatrix} 0 & 1 & 1\\ 1 & 0 & 1\\ 1 & 1 & 0 \end{pmatrix}

Find an orthogonal matrix $P$ and a diagonal matrix $D$ such that $A=PDP^T$. (Hint: The eigenvalues of $A$ are all integers.

I managed to do this by finding the eigenvalues through solving the characteristic equation, and then finding the corresponding eigenspace. After that I went ahead to use the Grand-Schdmit process, and got an orthnormal basis.

My solution was mighty complicated, involving 1.5 pages of workings.

Is there a faster way to do this? Question is worth 5 marks.

share|improve this question
    
It might be a good way to try to transform the matrix into a quadratic form, and put it into the sum of squares. Or else you could try the Gaussian elimination process, both on the right and on the left sides. –  awllower Nov 21 '12 at 12:33

3 Answers 3

Denote by $B$ the matrix $$ B = A + I = \begin{pmatrix} 1 & 1 & 1\\ 1 & 1 & 1\\ 1 & 1 & 1 \end{pmatrix}. $$ If you manage to diagonalize orthogonally $B$, then $PBP^T = D$ for some diagonal $D$ and then $PAP^T = D - I$ where $D - I$ is also diagonal. The eigenvalues of $B$ are $3, 0, 0$ which is best seen by writing directly the eigenvalue equation $(B - \lambda I)(x,y,z)^T = 0$: $$ (1 - \lambda)x + y + z = 0, \\ x + (1 - \lambda)y + z = 0, \\ x + y + (1 - \lambda)z = 0. $$ Adding the three equations you get: $$ (3 - \lambda)(x + y + z) = 0. $$ So either $\lambda = 3$, and for which the corresponding eigenvector is $(1,1,1)^T$ or $x + y + z = 0$, for which the corresponding eigenvalue is $0$ (with multiplicity two). The eigenvalues of $A$ will be $2, -1, -1$, find an orthogonal basis for $x + y + z = 0$, and you're done.

share|improve this answer

Using the Gram-Schmidt process, even if you spell it correctly, is not a great idea for this kind of question, as it will in general spoil the hard-earned property of being a basis of eigenvectors. Here is an opportunistic approach that works for this particular matrix $A$.

On thing that is easy to note is that $A+I$ has all entries $1$, so that not only its determinant is $0$, but its rank is only $1$: $$ A+I=\begin{pmatrix}1\\1\\1\end{pmatrix}\cdot\begin{pmatrix}1&1&1\end{pmatrix} $$ Thus the eigenspace for $\lambda=-1$ is the $2$-dimensional kernel of the right matrix $(1~~1~~1)$, the plane given by the equation $x+y+z=0$. Since the trace of $A$ is the sum of its eigenvalues (taken with multiplicities), and it is $0$, the remaining eigenvalue is $2$, and an eigenvector for it is $(1~~1~~1)^T$, which is orthogonal (as it should be given that $A$ is symmetric) to the other eigenspace $x+y+z=0$.

Now to find $P$ we need an orthogonal basis of the eigenspace $x+y+z=0$, and then normalise all those eigenvectors. I can't see any particularly nice orthogonal pairs in that eigenspace, so I'll just take the first vector that comes to mind $(1~~-1~~0)^T$ and take the cross product with the eigenvector $(1~~1~~1)^T$ for $\lambda=2$, to give $(-1~~-1~~2)^T$. Dividing by the norms of these vectors I find $$ P=\frac1{\sqrt6} \begin{pmatrix}\sqrt3&-1&\sqrt2\\-\sqrt3&-1&\sqrt2\\0&2&\sqrt2\end{pmatrix} \qquad \text{and}\qquad D=\begin{pmatrix}-1&0&0\\0&-1&0\\0&0&2\end{pmatrix}. $$

share|improve this answer

I suppose you have managed to show that the eigenvalues of $A$ are $\{2,-1,-1\}$ and the only eigenvector (up to scalar multiple) for $\lambda=2$ is $v=(1,1,1)^T$. The problem now is to find $P$ easily.

Note that for a real symmetric matrix, eigenvectors for different eigenvalues are orthogonal to each other. Therefore, if we can find an orthogonal matrix $P$ whose first column is parallel to $v$, then the other two columns will automatically become eigenvectors for $\lambda=-1$. Now, to find such a matrix $P$, we can use Householder reflection. The relevant formula is $$P = I - \frac{2uu^T}{\|u\|^2},\ \textrm{ where } u=v-\|v\|(1,0,0)^T.$$ Put $v=(1,1,1)^T$, we get $$ P = \frac1{\sqrt{3}}\begin{pmatrix} 1 & 1 & 1\\ 1 & {\small\frac{\sqrt{3}-1}{2}} & {\small\frac{-\sqrt{3}-1}{2}}\\ 1 & {\small\frac{-\sqrt{3}-1}{2}} & {\small\frac{\sqrt{3}-1}{2}} \end{pmatrix}. $$ You may verify that $P$ is indeed an orthogonal matrix (i.e. $P^TP=I$) and $A=P\,\ \mathrm{diag}(2,-1,-1)\ P^T$.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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