When looking at a quadratic form in a matrix, I am not completely sure how to tell if is one of the answers from above.

-positive (semi)definite

-negative (semi)definite


how do i figure this out


There are two approaches. One is to write the (symmetric) matrix associated the quadratic form, the other is to "complete the square", as the a deleted answer notes.

For example: $$ q(x_1,x_2) = x_1^2 - 4x_1x_2 + x_2^2 = x_1^2 - 2x_1x_2 - 2x_2x_1 + x_2^2 $$ The matrix associated with this quadratic form is $$ M = \pmatrix{1&-2\\-2&1} $$ and we wish to determine its definiteness. Applying Sylvester's criterion to both $M$ and $-M$ allows us to check if $M$ is positive/negative definite. Since the $2 \times 2$ determinant $\det(M)$ is negative, we can conclude that $M$ is neither positive nor negative definite.

Now, $\det(M) \neq 0$, so $M$ is invertible. Since $M$ is invertible without being positive/negative definite, it is invertible.

If $M$ were singular, we'd have to try something else. In practice (with pencil and paper), one would find the eigenvalues of $M$. If all non-zero eigenvalues have the same sign, then the singular $M$ is postive/negative semidefinite. Otherwise, $M$ would be indefinite.

Another approach to apply (especially for large matrices) is to attempt a Cholesky factorization. If $M$ has a Cholesky factorization, it is positive semidefinite. If $-M$ has a Cholesky factorization, it is negative semidefinite. Otherwise, $M$ is neither.

In this case, we find $M$ and $-M$ have no Cholesky factorization

Equivalent (in a sense) to Cholesky factorization is the approach of completing the square. We find $$ q(x_1,x_2) = x_1^2 - 4x_1x_2 + x_2^2 = (x_1^2 - 4x_1x_2 + 4x_2^2) - 3x_2^2 = \\ (x_1 - 2x_2)^2 - 3x_2^2 $$ Correspondingly, we see that $$ M = \pmatrix{1&0\\-2&1}^T\pmatrix{1&0\\0&-3}\pmatrix{1&0\\-2&1} $$


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.