0
$\begingroup$

Let $A$ be a symmetric positive definite matrix. I am interested in finding a general expression for the $A^{-1}$. Using the Cayley-Hamilton Theorem (https://en.wikipedia.org/wiki/Cayley%E2%80%93Hamilton_theorem), we have that $A^{-1}$ can be expressed as \begin{aligned}A^{-1}={\frac {(-1)^{n-1}}{\det A}}(A^{n-1}+c_{n-1}A^{n-2}+\cdots +c_{1}I_{n}),\end{aligned}

I am bit puzzled finding the coefficients $c_i$, when $A$ is symmetric and positive definite. Also, would it be possible to avoid evaluating the $det{A}$, that is finding a expression equivalent to it?

$\endgroup$
1
$\begingroup$

$$\det(\lambda I-A)=\lambda^n+\beta_1\lambda^{n-1}+...+\beta_n,$$ where $$\beta_i=(-1)^i( \text { the sum of the principal minors of A of order }i),1 \le i \le n.$$ Note that $\beta_n=(-1)^n\det(A).$The easiest way to evaluate a determinant, whether $\det(A)$or any other of the principal minors of $A$ is by pivotal condensation, which is automatic row-reduction.However, there is an alternate formula, due to Bocher, for evaluating the $\beta_i$ $$\beta_1=-\text {trace}(A),$$ $$\beta_i=\frac{-1}{i}[1,\beta_1,...,\beta_{i-1}]\bullet [\mu_i,\mu_{i-1},...,\mu_1],2 \le i \le n,$$ where $\mu_i=$trace$(A^i),1 \le i \le n.$ Note that the most efficient way to calculate such an expression as $$A^3+\beta_1 A^2+\beta_2A+\beta_3I$$ is by "synthetic division" $$A(A(A+\beta_1I)+\beta_2I)+\beta_3I$$ working outwards from the innermost parenthesis and calculating successive powers of $A$, the $\mu_i$ and the $\beta_i$ recursively. Bocher's formula is valid over any field of characteristic 0 or of characteristic > $n.$

$\endgroup$

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.