1
$\begingroup$

Suppose $A \in \mathbb{R}^{nxn}$ symmetric and positive definit.

We can show that: $$\exists \epsilon > 0 \in \mathbb{R}: \forall x \in \mathbb{R}^n: x^TAx \ge \epsilon \Vert x \Vert^2$$

My claim is that the smallest eigenvalue of $A$ is the supremum for those $\epsilon$.

I'd appreciate to see a prove/disprove!

$\endgroup$
3
$\begingroup$

Yes, this is true. Take diagonalization $A=TDT^t$ with $D=\operatorname{diag}(\lambda_1,\ldots \lambda_n)$. We obtain with $\lambda_0:=\min_{i=1..n}\lambda_i$ \begin{align} (Tx)^tATx&=x^tT^tATx=x^tDx=\sum_{i=1}^n \lambda_i x_i^2 \geq \lambda_0\sum_{i=1}^n x_i^2 = \lambda_0 \|x\|^2 = \lambda_0 x^tx \\&= \lambda_0 x^tT^tTx= \lambda_0 (Tx)^tTx = \lambda_0\|Tx\|^2. \end{align} By the substitution $x=T^ty$ follows $y^tAy\geq \lambda_0\|y\|^2$. Its clear that this inequality is sharp (take an eigenvector of $\lambda_0$ as y).

$\endgroup$
2
$\begingroup$

Let's solve the following minimization problem: $$ \begin{aligned} &\min x^TAx\\ &s.t.\ x^Tx=1. \end{aligned} $$ Using the Lagrange method: $$ L=x^TAx-\lambda(x^Tx-1). $$ So $$ \frac{\partial L}{\partial x}=x^T(A+A^T)-2\lambda x^T=2(Ax)^T-2\lambda x^T=0, $$ or $Ax=\lambda x$, and the minimum is reached at an eigenvector $x$. This immediately proves that indeed for any $\epsilon\le\lambda$ and for any $x$ the inequality $x^TAx\ge\epsilon x^Tx$ (and backward, if the inequality holds, then $\epsilon\le\lambda$).

Actually, we never used the fact that $A$ is positive definite. The statement holds for any symmetric matrix, just $\epsilon$ might be negative in that case.

$\endgroup$

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.