Given $Q(x,y,z;\alpha)=x^2+z^2+2\alpha xy+2xz$, i have to study the sign of quadratic form. Obviously i can use eigenvalues or studying the sign of minors, but in this case i have a hard time to understand how to act. In fact, for $A=\begin{bmatrix} 1 & \alpha & 1\\ \alpha & 0&0\\ 1& 0&1 \end{bmatrix}$ I have:


$A_2=\det \begin{bmatrix} 1&\alpha \\ \alpha & 0 \end{bmatrix}=-\alpha^2>0\Rightarrow \alpha^2<0$,

$A_3=\det \begin{bmatrix} 1& \alpha &1\\ \alpha &0 &0\\ 1 &0 &1 \end{bmatrix}=-\alpha^2>0\Rightarrow \alpha^2<0$

So, $\left\{\begin{matrix} 1>0 \\ \alpha^2<0 \\ \alpha^2<0 \end{matrix}\right.=\left\{\begin{matrix} 1>0 \\ \alpha^2<0 \end{matrix}\right.$. Now, since for $\alpha^2<0$ no solutions exist, how can I conclude? Is the matrix positive-definite or what?

  • $\begingroup$ Why do you impose $-\alpha^2 > 0$? The computation of $A_2$ tells you exactly that the upper-left $2 \times 2$ minor defines a quadratic form of signature $(1, 1)$ (or a degenerate, nonzero positive semidefinite quadratic form iff $\alpha = 0$). $\endgroup$ – Travis Jan 8 at 10:39
  • $\begingroup$ @Travis Thanks for your answer. I thought that $>0$ was the default sign from which to start the analysis. But if not, what I have to impose for $-\alpha^2$? $\endgroup$ – Marco Pittella Jan 8 at 10:54
  • $\begingroup$ There's no notion of "default sign", and there's nothing to impose. Rather, $-\alpha^2$ controls the signature of the resulting form. $\endgroup$ – Travis Jan 8 at 11:21
  • 1
    $\begingroup$ Note, too, by the way, that when determining signature using eigenvalues, you don't actually need to determine the eigenvalues themselves (which in this case would involve solving a cubic), you only need their signs, and you can extract that information from Descartes' Rule of Signs. If it would be useful, I'm happy to write up a short solution explaining this method. $\endgroup$ – Travis Jan 8 at 11:23
  • $\begingroup$ @Travis Thanks again for your help. $-\alpha^2$ controls the sign of the form, of course. So, how can I conclude in this case? Any value I put on place of $\alpha$ I obtain an indefinite form. Could you show me the steps to follow? Instead, for the method that you mentioned well, ok, thank you so much for your time. $\endgroup$ – Marco Pittella Jan 8 at 11:40

Following a remark in the comments, here's an efficient way to calculate the sign using the characteristic polynomial---recall that the signature is essentially a count of the positive, zero negative eigenvalues of the representative matrix---but without actually computing its roots.

The characteristic polynomial of the matrix representation $A$ of the quadratic form is $$\det \left(t I_3 - \pmatrix{1&\alpha&1\\ \alpha&0&0\\1&0&1}\right) = t^3 - 2 t^2 - \alpha^2 t + \alpha^2 .$$ For $\alpha \neq 0$, we have $\alpha^2 > 0$, in which case the number of sign changes of the coefficients is $2$, and thus by Descartes' Rule of Signs (and the fact that all of the eigenvalues of a real, symmetric matrix are real) the polynomial has (1) two positive roots, and, (2) since $0$ is not a root, one negative root. Thus, the signature is $(2, 1)$ (Lorentzian). NB we avoided computing explicitly the roots of the characteristic polynomial, which would have been difficult.

On the other hand, if $\alpha = 0$, the constant term of the characteristic polynomial is zero, so in that case the quadratic form is degenerate.


Here, it is an elementary solution. After completing the square $$ Q(x,y,z)=(x+\alpha y +z)^2 -\alpha^2(y+\frac{1}{\alpha}z)^2+z^2 $$ assuming $\alpha\neq 0$. It means that $Q$ indefinite quadratic form if $\alpha\neq 0$, otherwise it is positive semi definite only.


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.