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.

This question is very similar to this one.

Let $B$ be a positive semi-definite matrix and $B = \begin{bmatrix} B_{11} & B_{12} \\ B_{12}' & B_{22} \end{bmatrix}$ where $B_{11}$ is $p \times p$. then $\lambda_1(B) \le \lambda_1(B_{11}) + \lambda_1(B_{22})$ where $\lambda_1$ is the largest eigenvalue of the matrix in argument.

How can I show it using extremal representation of the maximum eigenvalue of a symmetric matrix.

$$ \lambda_1(B) = max_{||x||=1} x'Bx \\ = max_{||x||=1} [x_1' x_2']\begin{bmatrix} B_{11} & B_{12} \\ B_{12}' & B_{22}\end{bmatrix} \begin{bmatrix} x_1 \\ x_2 \end{bmatrix} \\ = max_{||x||=1} [x_1' x_2']\begin{bmatrix} B_{11}x_1 + B_{12}x_2 \\ B_{12}'x_1 + B_{22}x_2\end{bmatrix} \\ = max_{||x||=1} { x_1'B_{11}x_1 + x_1'B_{12}x_2 + x_2'B_{12}'x_1 + x_2'B_{22}x_2 } $$

If $x_1'B_{12}x_2 + x_2'B_{12}'x_1 $ can go to zero than I am done. But this is certainly not the case.

Where am I making a mistake?

share|improve this question

1 Answer 1

up vote 4 down vote accepted

Write $x=c_1u_1+c_2u_2$ with unit vectors $u_1$ and $u_2$ that are non-zero only in the first and second block, respectively. Then

$$x'Bx=c'Ac$$

with

$$A=\pmatrix{u_1'Bu_1&u_1'Bu_2\\u_1'Bu_2&u_2'Bu_2}$$

symmetric and positive-(semi)definite. Since $u_1'Bu_1\le\lambda_1(B_{11})$ and $u_2'Bu_2\le\lambda_1(B_{22})$, we have $\operatorname{tr}A\le\lambda_1(B_{11})+\lambda_1(B_{22})$. Since the eigenvalues of $A$ are non-negative, this implies $\lambda_1(A)\le\lambda_1(A)+\lambda_2(A)=\operatorname{tr}A\le\lambda_1(B_{11})+\lambda_1(B_{22})$. Since any unit vector $x$ can be written in the above form with $c$ a unit vector, $x'Bx=c'Ac\le\lambda_1(B_{11})+\lambda_1(B_{22})$ for $x$ a unit vector, so

$$\lambda_1(B)=\max_{\lVert x\rVert=1}x'Bx\le\lambda_1(B_{11})+\lambda_1(B_{22})\;.$$

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.