Determinant of a 2x2 Quadratic Form
Suppose we have a 2 quadratic form . How can we compute its determinant?
This problem is a little under-specified. In fact, we can always pick a basis for so that our quadratic form is diagonal with diagonal entries or . So there is always some basis where our quadratic form has determinant or .
But usually, we have a choice of inner product on , and if we restrict to orthonormal bases with respect to this inner product then does indeed have well-defined eigenvectors, and a well-defined trace and determinant. And once we have the eigenvectors, the trace and determinant are easy to evaluate. If are the eigenvectors with eigenvalues , then the trace of is given by and the determinant of is given by
All this eigenvector business is actually not needed for computing the trace: we can pick any orthonormal vectors , and the trace is given by .
However, the determinant is trickier. In general, is not equal to for a general pair of orthonormal vectors and . To see what goes wrong, we can write out as a matrix in the basis. The diagonal entries of this matrix are simply and . To obtain the off-diagonal entries, we can apply the polariziation identity (essentially a fancy name for the fact that ). Now that we have the matrix entries for , we can apply the usual formula that and we find that
This formula simplifies a little if we define a new unit vector which lies half way in between and . Then we can do a little bit of algebra to find that
We can simplify this formula even more by introducing the complementary unit vector . Note that also form an orthonormal basis for . Since we can use the any orthonormal basis of to compute the trace of , we know that , and therefore the expression in our formula is simply equal to . So in the end, the determinant of is given by where and form one orthonormal basis, and and form another orthonormal basis rotated by .
As one last sanity check, note that if we take and to be the eigenvectors and , then , and we so find again that the determinant is given by . But now we see that this is really a special case, and in general we need to add a correction term involving a pair of vectors rotated by from our starting vectors.
No comments: