Expressing a symmetric matrix in terms on eigenvalues/vectors

1. May 9, 2013

han35

1. The problem statement, all variables and given/known data
Generate a random 10 x 10 symmetric matrix A (already done in MATLAB) . Express A in the form

2. Relevant equations

$A =$ $\displaystyle \sum_{j=1}^{10}λ_j(A)v_jv^T_j\$

for some real vectors $v_j, j = 1, 2, . . . , 10.$

3. The attempt at a solution

I'm pretty sure the solution has something to do with the eigenvectors of a symmetric matrix being orthogonal.

The whole sum is basically $A = λ_1v_1v^T_1 + λ_2v_2v^T_2 . . . λ_{10}v_{10}v^T_{10}.$

If $v_j$ are an eigenvectors then we can express that as:

$A = Av_1v^T_1 + Av_2v^T_2 . . . Av_{10}v^T_{10}$
$→ A = A( v_1v^T_1 + v_2v^T_2 . . . v_{10}v^T_{10} )$

But that means we need $v_1v^T_1 + v_2v^T_2 . . . v_{10}v^T_{10} = I$. But I don't know if these vectors are meant to have those properties. =S

Basically I need to know what the $v_j$'s are

2. May 9, 2013

han35

Ok, so after testing some symmetric 3x3 matrices and computing it by hand I can confirm that the $v_j$'s are indeed the eigenvectors of $A$.

Now I just need to know why, lol.

3. May 10, 2013

han35

Does anyone know why?

If A is a symmetric matrix then A can be expressed as:

$A =$ $\displaystyle \sum_{j=1}^{n}λ_j(A)v_jv^T_j\$

where $v_j, j = 1, 2, . . . , n.$ are the eigenvectors of $A$

But why???

4. May 10, 2013

Dick

Your $v_j$ are an orthonormal set of eigenvectors, right? Multiply $sum_{j=1}^{n}λ_j(A)v_jv^T_j$ by any eigenvector $v_k$. You get the same thing as multiplying that vector by A, right? If you want a fancy name for it, it's a simple version of the spectral theorem.