Given two symmetric positive definite matrices $A,B\in\mathbb{R}^{n\times n}$ and $x\in\mathbb{R}^n$.

How do I prove that the generalized Rayleigh quotient

$R(A,B,x):=\dfrac{x\cdot A\cdot x}{x\cdot B\cdot x}$

has a maximum $\lambda_{max}$, which corresponds to the largest eigenvalue of the generalized eigenvalue problem

$A\cdot x=\lambda\cdot B\cdot x$?

I have tried to complete the proof by substituting $B=C^2$ and $y=C\cdot x$ to get

$\dfrac{x\cdot A\cdot x}{x\cdot B\cdot x}=\dfrac{y\cdot C^{-T}\cdot A\cdot C^{-1}\cdot y}{y\cdot y}$.

Defining $D:=C^{-T}\cdot A\cdot C^{-1}$ we have $D=D^T$ and then it is simple to show that the maximum eigenvalue of $D$ is the maximum of the quotient by substituting $D=\sum\limits_j\lambda_jv_j\otimes v_j$ and $y=\sum\limits_jy_jv_j$ into the quotient. I can, however, not seem to prove that the eigenvalues of $D$ from

$D\cdot v_j=\lambda_jv_j\quad\Leftrightarrow\quad C^{-T}\cdot A\cdot C^{-1}\cdot v_j=\lambda_jv_j\quad\Leftrightarrow\quad A\cdot C^{-1}\cdot v_j=\lambda_j\cdot C^T\cdot v_j$

correspond to the eigenvalues of

$A\cdot v_j=\lambda_j\cdot C^2\cdot v_j\quad\Leftrightarrow\quad A\cdot v_j=\lambda_j\cdot B\cdot v_j$.


Setting $w_j = C^{-1} v_j$, we have $$ A(C^{-1}v_j) = \lambda_j C^T v_j \iff\\ A(w_j) = \lambda_j (C^T C) w_j $$ Making sure that you've chosen $C$ to be normal (verify that such a choice always exists), we note that $C^TC$ and $C$ share eigenvectors.

  • $\begingroup$ I believe this answer suffices to complete the proof. Your last remark is useful, but I believe it is not even necessary to complete the proof, since the eigenvalues $\lambda_j$ do not change under the substitution $w_j:=C^{-1}\cdot v_j$. And thus, $\lambda_{max}$ is still the largest eigenvalue of the generalized eigenproblem $A\cdot x=\lambda\cdot B\cdot x$. $\endgroup$ – Adriaan Nov 11 '14 at 9:41
  • $\begingroup$ Ah, you're right. I was still thinking in terms of the eigenvalues of $A$. $\endgroup$ – Omnomnomnom Nov 11 '14 at 12:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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