Assume we have a matrix $A$ that satisfies this relation $Ax=b$ $$A\begin{bmatrix} 1\\0\\1 \end{bmatrix}=\begin{bmatrix} 1\\3\\-2\end{bmatrix}$$ Is $A$ postive definite, positive semidefinite, negative definite, negative semidefinite, or indefinite? How would we find out this information given any $\vec x$ and $\vec b$?

  • $\begingroup$ There are many equivalent definitions, but a useful one here is that a matrix $A$ is positive definite iff $|\vec x|>0\implies\vec x^tA\vec x>0\ \forall\ \vec x$. If you know $A\vec x = \vec b$, then $\vec x^tA\vec x = \vec x\cdot\vec b$. Checking a finite number of vectors won't necessarily suffice,though. $\endgroup$ – Kajelad Aug 11 '17 at 20:52
  • $\begingroup$ Is there an algorithm we can use to determine the definiteness? What about a way to directly prove that it is indefinite? $\endgroup$ – D.R. Aug 11 '17 at 21:07
  • $\begingroup$ Can you find an example of an $A$ that fits your criteria? What are its eigenvalues? $\endgroup$ – Doug M Aug 11 '17 at 21:57
  • $\begingroup$ I think MathLover's answer is the only conclusion we can draw from a single equation. If we have 3 linearly independent values for $\vec x$ and corresponding values for $\vec b$, we can solve the matrix equation $AX=B$ for $A$ and determine the eigenvalues directly. $\endgroup$ – Kajelad Aug 11 '17 at 23:12

In general, you want to determine the definiteness of a $n\times n$ matrix $A$ given vectors $\vec x$, $\vec b$ such that $A\vec x=\vec b$. Just to clarify, I'll define definiteness as follows:

$A$ is positive definite $\implies \vec x^tA\vec x>0\ \forall \vec x\neq \vec 0$

$A$ is negative definite $\implies \vec x^tA\vec x<0\ \forall \vec x\neq \vec 0$

$A$ is positive semidefinite $\implies \vec x^tA\vec x\ge 0\ \forall \vec x\neq \vec 0$

$A$ is negative semidefinite $\implies \vec x^tA\vec x\le0\ \forall \vec x\neq \vec 0$

$A$ is indefinite if none of the above are true.

Since $\vec x^TA\vec x=\vec x\cdot \vec b$, we can evaluate the above inequality once for each value for $\vec x$ and $\vec b$, but only one set of values is insufficient to completely determine the definiteness of $A$ (except in the one dimensional case). However, if $\vec x\cdot\vec b$ is a counterexample to any of the above inequalities, we can of course conclude that $A$ does not have the corresponding definiteness.

If we want to check the inequalities above for all $\vec x$, it's sufficient (and, in fact, necessary) to have a set of vectors that form a basis for the entire space, since we can write any $\vec x^tA\vec x$ in terms of this basis. The simplest way to do this is to completely determine $A$.

Suppose we know a linearly independent set $\{\vec x_1,\vec x_2,...,\vec x_n\}\subset\mathbb R^n$, as well as another set $\{\vec b_1,\vec b_2,...,\vec b_n\}\subset\mathbb R^n$, such that $A\vec x_i=\vec b_i\ \forall\ i$. We can then let $X$ and $B$ be matrices whose columns are $\vec x_i$ and $\vec b_i$ respectively, giving us the following matrix relation.


Since $X$ has linearly independent columns, we know its inverse must exist. Thus we can compute $A$.


Definiteness is typically only discussed for Symmetric matrices, so if $A$ is not symmetric, it may be preferred to consider its symmetric part $A_{sym}=\frac 12(A+A^T)$, which defines the same quadratic form.

Once we know $A$ we can determine its definiteness several ways. One is to find all of its eigenvalues. If $A$ is symmetric we can find the determinants of the principal minors and use Sylvester's Criterion.


This matrix is certainly not positive definite or positive semidefinite as $\begin{bmatrix}1&0& 1\end{bmatrix} A \begin{bmatrix}1\\0\\ 1\end{bmatrix}=-1 < 0$.


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.