I have found an exercise involving a $3\times 3$ projection matrix which projects to a space $V$ with $\mathrm{dim}(V)=2$.
The matrix(or operator) is defined as $P(x)=v*(x, v)+u*(x, u)$.
So, in my understanding it projects vectors $x$ onto two vectors $u$ and $v$.
Now, the exercise asked us to find it's eigenvalues. So, I found three of them and each one belongs to a different eigenvector(the set of the eigenvectors is orthogonal).
So, upon solving the exercise, I found myself really not understanding this conceptually.

Why would there be 3 orthogonal eigenvectors which span a space of dimension equal to 3 when the projector projects to a space of dimension equal to 2?

EDIT: This is the exercise:
$\def\Bra#1{\left\langle#1\right|}\def\Ket#1{\left|#1\right\rangle}\def\Braket#1{\mathinner{\left\langle{#1}\right\rangle}}$ We have an orthonormal basis $\{\Ket{1},\Ket{2},\Ket{3}\}$. We also have the vectors $\Ket{\Psi}=\Ket{1}-\Ket{2}+\Ket{3}$ and $\Ket{\Phi}=\Ket{1}+\Ket{2}$.
We define an operator(matrix) as: $Q=\Ket{\Psi}\Bra{\Psi}+\Ket{\Phi}\Bra{\Phi}$ with $\Bra{\Psi}$ and $\Bra{\Phi}$ being the complex transpose of each vector.

Solution: i find $\Ket{\Psi}=(1, -1 , 1)^T$ and $\Ket{\Phi}=(1, 1, 0)^T$ with inner products $\Braket{\Psi\mid\Psi}=3$, $\Braket{\Phi\mid\Phi}=2$.
To find the eigenvalues of $Q$ i just act with $Q$ of $\Ket{\Psi}$: $Q\Ket{\Psi}=\Ket{\Psi}\Braket{\Psi\mid\Psi}+\Ket{\Phi}\Braket{\Phi\mid\Psi}=3\Ket{\Psi}$ because $\Ket{\Psi}$ and $\Ket{\Phi}$ are orthogonal.
Also i act with $Q$ on $\Ket{\Phi}$: $Q\Ket{\Phi}=\Ket{\Psi}\Braket{\Psi\mid\Phi}+\Ket{\Phi}\Braket{\Phi\mid\Phi}=2\Ket{\Phi}$.
So we have two eigenvalues: $\lambda=2$ with eigenvector $\Ket{\Phi}$ and $\lambda=3$ with eigenvector $\Ket{\Psi}$.
The solution says that there must be another vector orthogonal to $\Ket{\Phi}$ and $\Ket{\Psi}$ because it is a three dimensional space. That eigenvalue is zero with eigenvector(i have computed it) $(-1,1,2)^T$.

  • $\begingroup$ Are you sure that all three eigenvectors are associated with the same eigenvalue? $\endgroup$ – amd Apr 30 '16 at 16:33
  • $\begingroup$ @amd "I found three of them and each one belongs to a different eigenvector". No, each eigenvalue belongs to its own eigenvector. $\endgroup$ – TheQuantumMan Apr 30 '16 at 17:16
  • $\begingroup$ Any projection has only two eigenvalues. If you found a third, you did something wrong. $\endgroup$ – amd May 1 '16 at 6:10
  • $\begingroup$ @amd no, the solution of the exercise gives three eigenvalues. It's a 3x3 matrix. Wouldn't we expect to find 3? $\endgroup$ – TheQuantumMan May 1 '16 at 9:49
  • $\begingroup$ I’d expect at most three. For any projection, the only eigenvalues are $0$ and $1$, with multiplicities $>1$ for a $3\times 3$ or larger matrix. $\endgroup$ – amd May 2 '16 at 6:06

Any non-degenerate projection $P$ onto a (sub)space $V$ has only two eigenvalues: $1$, with associated eigenspace $V$ itself, and $0$, with associated eigenspace $\ker(P)$. In this case, $P$ is an orthogonal projection, so it should come as no surprise that eigenvectors associated with $0$ are orthogonal to those associated with $1$. After all, $P$’s kernel is the orthogonal complement of $V$. Since all elements of $V$ are eigenvectors of $P$ (including, not surprisingly, $u$ and $v$ themselves), you can certainly choose an orthogonal pair of them that span $V$ and thus end up with an orthogonal basis for the entire space that consists of eigenvectors of $P$.

Update: That being said, the operator $Q$ in the exercise isn’t, strictly speaking, a projection since $Q^2\ne Q$. The vectors $\Psi$ and $\Phi$ would have to be unit vectors for $Q$ to be a projection in this sense. Nevertheless, it does map the three-dimensional domain onto the two-dimensional span of $\Psi$ and $\Phi$, so you can expect it to have $0$ as an eigenvalue with the kernel of $Q$ as the associated eigenspace. By construction of $Q$, this space will consist of all vectors that are orthogonal to both $\Psi$ and $\Phi$, i.e., it is the orthogonal complement of their span. Moreover, eigenvectors associated with other eigenvalues must lie in the span of $\Psi$ and $\Phi$, and so will be orthogonal to this eigenspace as well.

  • $\begingroup$ So, because the third eigenvector corresponds to an eigenvalue=0 it is a basis for null(P) while V=range(P)(the image) so it is not surprising. I thought that all three eigenvectors form a basis for V. So, it was a misunderstanding, i forgot that eigenvectors with eigenvalue=0 are not a basis for range(V) but for null(P). $\endgroup$ – TheQuantumMan May 2 '16 at 10:03

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.