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.

If $\lambda$ is the largest eigenvalue of a real symmetric $n \times n$ matrix $H$, how can I show that: $$\forall v \in \mathbb{R^n}, ||v||=1 \implies v^tHv\leq \lambda$$

Thank you.

share|improve this question
    
add comment

3 Answers

Hint:

Real symmetric matrices are diagonalizable.

Hint 2 (added after reading comments on posts):

A matrix is diagonalizable by a suitable choice of coordinates if and only if there is an eigenbasis. (taken from here)

share|improve this answer
    
The matrix $\begin{bmatrix}1& 2 \\0 &0\end{bmatrix}$ is diagonalizable. You want to you orthogonal diagonalization. en.wikipedia.org/wiki/Symmetric_matrix#Properties –  Jonas Meyer Mar 29 '11 at 12:03
    
ok, so $H$ is diagonalizable by an orthogonal matrix. Is the spectral theorem somehow related here? –  user8837 Mar 29 '11 at 13:19
    
@Rafael: According to Wikipedia, the spectral theorem gives you conditions under which a matrix is diagonalizable, so yes, I think the spectral theorem is related. –  Matt N. Mar 29 '11 at 14:13
add comment

Step 1: All Real Symmetric Matrices can be diagonalized in the form: $ H = Q\Lambda Q^T $ So, $ {\bf v}^TH{\bf v} = {\bf v}^TQ\Lambda Q^T{\bf v} $

Step 2: Define transformed vector: ${\bf y} = Q^T{\bf v}$.

So, ${\bf v}^TH{\bf v} = {\bf y}^T\Lambda {\bf y} $

Step 3: Expand

${\bf y}^T\Lambda {\bf y} = \lambda_{\max}y_1^2 + \lambda_{2}y_2^2 + \cdots + \lambda_{\min}y_2^2 $

\begin{eqnarray} \lambda_{\max}y_1^2 + \lambda_{2}y_2^2 + \cdots + \lambda_{\min}y_N^2& \le & \lambda_{\max}y_1^2 + \lambda_{\max}y_2^2 + \cdots + \lambda_{\max}y_N^2 \\ & & =\lambda_{\max}(y_1^2 +y_2^2 + \cdots y_N^2) \\ & & =\lambda_{\max} {\bf y}^T{\bf y} \\ \implies {\bf y}^T\Lambda {\bf y} & \le & \lambda_{\max} {\bf y}^T{\bf y} \end{eqnarray}

Step 5: Since $Q^{-1} = Q^T, QQ^T = I $ \begin{eqnarray} {\bf y}^T{\bf y} &= &{\bf v}^TQQ^T{\bf v} = {\bf v}^T{\bf v} \end{eqnarray}

Step 6: Putting it all back together \begin{eqnarray} {\bf y}^T\Lambda {\bf y} & \le & \lambda_{\max} {\bf y}^T{\bf y} \\ {\bf v}^TH{\bf v} & \le & \lambda_{\max}{\bf v}^T{\bf v} \end{eqnarray}

By definition, $ {\bf v}^T{\bf v} = \|{\bf v}\|^2 $ and by definition $\|{\bf v}\| = 1$ \begin{eqnarray} {\bf v}^TH{\bf v} & \le & \lambda_{\max} \end{eqnarray} Boom!

share|improve this answer
add comment

Another hint along the same lines as Matt's: for which $\vec{v}$ is the LHS of your inequality maximised?

share|improve this answer
    
not sure I understand... if $x$ is an eigenvector of $\lambda$ then $Hx=\lambda x$ and $x^tHx=x^t\lambda x=\lambda ||x||$ –  user8837 Mar 29 '11 at 12:56
    
You have a constraint on the $\mathcal{l}_{2}$ norm in your original question, so if $x$ is an eigenvector of $H$ s.t. $\|x\|=1$ what does this give you? You're there anyway I think. –  Bob Durrant Mar 29 '11 at 13:30
    
FWIW, this is the more typical form of what you are trying to prove: Rayleigh quotient –  Bob Durrant Mar 29 '11 at 13:32
    
then $x^tHx=x^t\lambda x=\lambda$ but this is only for eigenvectors of $\lambda$, and I need to show this is true for all $v\in \mathbb{R^n}$ –  user8837 Mar 29 '11 at 13:38
    
Okay, but you also know there is an orthonormal basis for $\mathbb{R}^n$ with the basis vectors the unit eigenvectors of $H$ (okay?). So you can rewrite $v$ as...? –  Bob Durrant Mar 29 '11 at 13:47
show 5 more comments

Your Answer

 
discard

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