Lets say I want to minimize the inverse of the following matrix: $$A = \begin{bmatrix} \lambda_1 &0 & 0\\ 0& \lambda_2 &0\\0& 0& \lambda_3 \end{bmatrix}$$ where all $\lambda \geq 1$. I also read that the determinant of the a the matrix $A$ can be given by: $$\det(A) = \lambda_1 \lambda_2 \lambda_3 $$. The trace of inverse of $A$ is given by: $$trace(A^{-1}) = \frac{1}{\lambda_1}+\frac{1}{\lambda_2}+\frac{1}{\lambda_3}=\frac{\lambda_1 \lambda_2 + \lambda_2 \lambda_3 + \lambda_1 \lambda_3}{\lambda_1 \lambda_2 \lambda_3}$$

So for $\lambda \geq 1$, is it safe to say that minimizing the trace of inverse of of A is the same as maximizing the determinant of A?

  • $\begingroup$ What about the other entries ? Is A a diagonal matrix ? Or are just the entries below the diagonal zeros ? Please be more precise! $\endgroup$ – Peter May 8 '17 at 17:19
  • $\begingroup$ @RobertIsrael Trace of inverse $\endgroup$ – Anon123 May 8 '17 at 17:22
  • $\begingroup$ Minimizing or maximizing subject to what constraints? If your only constraint is $\lambda \ge 1$, then the determinant can be arbitrarily large, and the trace of $A^{-1}$ can be arbitrarily close to $0$. $\endgroup$ – Robert Israel May 8 '17 at 17:23
  • $\begingroup$ @Peter It is a diagonal matrix. In general trace is always the sum of eigenvalues, no? $\endgroup$ – Anon123 May 8 '17 at 17:23
  • $\begingroup$ I think investigating matrix logarithm may be more fruitful. Then also starting from $I$ ($\lambda_k = 1 , \forall k$) becomes more natural (why?). $\endgroup$ – mathreadler May 8 '17 at 17:27

This is not equivalent. Let $A_1 = diag(1,2,2)$ and $A_2 = diag(1,1,5)$. Then $$ trace(A_1^{-1}) = 2 < trace(A_2^{-1}), \quad det(A_1) = 4 < det(A_2) \, . $$


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.