Positive Definiteness Of Real Symmetric Non-singular Matrix

840 Views Asked by At

For every $4 \times 4$ real symmetric non-singular matrix $A$, there exists a positive integer $p$ such that :
A. $pI+A$ is positive definite.
B. $A^p$ is positive definite.
C. $A^{-p}$ is positive definite.
D. $\mathrm exp(pA)-I$ is positive definite.

My approach to the problem.
Since we know that a matrix $A \in \mathbb M_n(\mathbb R)$ is positive definite if it is symmetric and all its eigenvalues are strictly positive.

For (A), I have shown $pI+A$ is symmetric as $A$ is symmetric. Also, it is clear that all eigenvalues of $pI+A$ will be of the form $p+ \lambda _i$, where $\lambda_i$ are non-zero eigenvalues of matrix $A$, as $A$ is non-singular. Therefore, I can chose $p$ in such a way that all $p+ \lambda_i \gt 0$.

For (B), it is clear that $A^p$ is symmetric and, I can choose $p$ to be even that will show the positive definiteness for $A^p$.

For (C), same $p$ can be taken as in (B).

My question is how to show that (D) is symmetric or not, and if so how to define the eigenvalues for (d).

1

There are 1 best solutions below

4
On BEST ANSWER

By the spectral theorem we can diagonalize $A$ as $A=Q\Lambda Q^T$. It is a property of the matrix exponential that $$ \exp(pQ\Lambda Q^T) = Q\exp(p\Lambda)Q^T $$ And the matrix exponential of a diagonal matrix is just the ordinary exponential of its elements. On $(0,\infty)$ $e^x>1$. Can you see how to finish the proof?


Here's a more direct way to do it. Let $x \neq 0$ have norm $1$. Then $$ x^T(e^{pA} - I)x = x^T\left(\sum_{k=0}^\infty \frac {p^k}{k!}A^k - I \right)x $$ $$ = x^T\left(\sum_{k=1}^\infty \frac {p^k}{k!}A^k\right)x = \sum_{k=1}^\infty \frac {p^k}{k!} x^T A^k x. $$ Each $A^k$ is PD and $p>0$ so we have a sum of positive things which is positive.

If you're worried about convergence, note that $$ \lambda_{\max}(A^k) = \left(\sup_{x : ||x||=1} x^TAx\right)^k \in (0, \infty) $$ so $$ 0 < \frac {p^k}{k!} x^T A^k x \leq \frac{(p\lambda_{\max}(A))^k}{k!} $$ so the partial sums of $x^T(e^{pA}-I)x$ give a monotonic series bounded above by $e^{p\lambda_{\max}(A)}-1$, and therefore it converges.