Given $$Q(x_1,x_2) = (4+k)x_1^2 + 2kx_1x_2 + 2x_2^2$$ I'm asked to find the values of $k$ such that $Q$ be a positive semi-definite form. Now, to do so I've decided to find its characteristic polynomial $$P(\lambda) = \lambda^2 - k^2+ 6\lambda - k\lambda + 8 + 2k$$ in order to build this system of equations: $$\begin{cases} P(\lambda) = 0 \\ \lambda \geq 0 \end{cases}$$ that later leads me to a dead end or nonsensical conclusions as $$k \geq \frac{1}{2}(2 - \lambda - \sqrt{36 - 28 \lambda + 5 \lambda^2}) - \lambda $$

My biggest concern, apart from the fact that my result seems so stupid and not in accordance to what I'm given as the right answer, is that the $\lambda \geq 0$ in my system of equations seems abusive, since I have the impression it imposes that there is one and only one eigenvalue. I know that my system of equations stinks, but I don't know how to put the problem.

How should I procede to solve this exercice and, more generally, what method should I use to face this kind of problems?

Thank you in advance.


I found the answer to this exercice by solving the following equation, involving the coefficient matrix of the quadratic form: $$ \begin{vmatrix}4+k&k\\k&2\end{vmatrix} =-k^2+2k+8 = 0$$ leading to $$k =\begin{cases}\ \ \ 4\\-2\end{cases}$$ which is the answer given as correct. Nevertheless, is this the or a generally correct method?

  • $\begingroup$ The characteristic polynomial of a matrix is invariant under the transformations $M\mapsto PMP^{-1}$ (base change as an endomorphism), but not under $M\mapsto PMP^t$ (base change as a bilinear form), so it is not adapted to the study of quadratic forms. $\endgroup$ – Captain Lama Apr 3 '19 at 19:45
  • $\begingroup$ In other words, you are confusing diagonalization of an endormorphism with diagonalization of a quadratic form. $\endgroup$ – Captain Lama Apr 3 '19 at 19:46
  • $\begingroup$ The how should I procede, @Captain Lama? $\endgroup$ – Hal Apr 3 '19 at 19:47

See Sylvester's Law of Inertia

$$ \left( \begin{array}{rr} 1 & - \frac{k}{2} \\ 0 & 1 \end{array} \right) \left( \begin{array}{cc} 4+k & k \\ k & 2 \end{array} \right) \left( \begin{array}{rr} 1 & 0 \\ - \frac{k}{2} & 1 \end{array} \right) = \left( \begin{array}{cc} 4 + k - \frac{k^2}{2} & 0 \\ 0 & 2 \end{array} \right) $$

| cite | improve this answer | |
  • $\begingroup$ I'd be thankful if you told me how to find the $\textbf{S}$ matrix in such $\textbf{SQS}^T$, or redirected me to a method that allows the finding. $\endgroup$ – Hal Apr 4 '19 at 11:42
  • 1
    $\begingroup$ @toritoverdejo math.stackexchange.com/questions/1388421/… $\endgroup$ – Will Jagy Apr 4 '19 at 17:26

For $Q$ to be semidefinite, one of the eigenvalues of its matrix must be zero, i.e., it must be singular, which is why finding where its determinant vanishes as in your answer gives you possible values of $k$. You still have to verify that $Q$ is indeed positive-definite, and not negative-definite, for these values of $k$.

Relating this to your approach, for $Q$ to be positive-definite, one of the roots of the characteristic polynomial must be $0$. This gives you two equations to solve for $k$, specifically, $$\frac12(k+6\pm\sqrt{5k^2+4k+4})=0.$$ One of these has no real solutions and the solutions to the other match the ones from the determinant.

| cite | improve this answer | |

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.