Suppose we have a $3$-dim vector space $V$ and we choose any non orthogonal set of basis ${v_1,v_2,v_3}$. Now we consider the linear transformation that projects any vector in $V$ to span$(v_1,v_2)$. Clearly the matrix representation for this transformation is $$\begin{bmatrix} 1 & 0 & 0\\ 0 & 1 & 0\\ 0 & 0 & 0 \end{bmatrix}$$. The eigenvectors corresponding to eigen-value $1$ are ${v_1,v_2}$, and corresponding to $0$ is $v_3$. As the matrix is symmetric, it should have orthogonal eigenvectors. But $\langle v_1,v_3\rangle$ should be non zero as they are non orthogonal. Why is this?

  • 1
    $\begingroup$ The matrix you have written is with respect to the basis $\{v_1,v_2,v_3\}$. That is all. The eigenvectors for the matrix you have written is $e_1,e_2$ and $e_3$. With $e_1,e_2$ corresponding to eigenvalue 1 and $e_3$ corresponding to 0. $\endgroup$
    – Soby
    Mar 15 '21 at 6:26
  • 1
    $\begingroup$ The matrix you have written down represents the transformation with respect to the basis $v_1,v_2,v_3$. With respect to that basis, $v_1$ is $(1,0,0)$ and $v_3$ is $(0,0,1)$, and those are orthogonal. (@Soby beat me by a few seconds!) $\endgroup$ Mar 15 '21 at 6:28
  • $\begingroup$ @Gerry Myerson , so we just consider the coordinate space in this which has standard basis and those are orthonormal? Can I say any symmetric matrix w.r.t non orthogonal basis has orthonormal eigenvectors in the coordinate space, but when I see in terms of the actual basis, it may not be so? I am confused when do we consider the coordinate space and when the actual vector space $\endgroup$ Mar 15 '21 at 6:46
  • $\begingroup$ Thanks @MorganRodgers so this is where I made a silly mistake, this arose my confusion.. $\endgroup$ Mar 15 '21 at 7:44
  • $\begingroup$ Yes this clears the picture thanks all.. $\endgroup$ Mar 15 '21 at 7:49

You're trying to invoke the following result:

Suppose $A$ is a symmetric real matrix (or a Hermitian complex matrix, if you prefer). Then the eigenspaces of $A$ are orthogonal.

This is true. The matrix you provided has orthogonal eigenspaces $\operatorname{span}\{(1, 0, 0), (0, 1, 0)\}$ and $\operatorname{span}\{(0, 0, 1)\}$. Note that these are subspaces of $\Bbb{R}^3$, which may or may not have anything to do with the inner product space $V$. Thus, it is not correct to say that these are eigenspaces of your projection operator $P$.

If you want the equivalent to the above result for general inner product spaces, this is it:

Suppose $V$ is a finite-dimensional inner product space, and $T : V \to V$ is self-adjoint. Then the eigenspaces of $T$ are orthogonal.

In our case, the projection map $$P : a_1 v_1 + a_2 v_2 + a_3 v_3 \mapsto a_1 v_1 + a_2 v_2$$ is not self-adjoint if $v_3$ is not orthogonal to $v_1$ and $v_2$. Let's suppose it's not orthogonal to $v_1$. Then $$\langle Pv_1, v_3 \rangle = \langle v_1, v_3 \rangle \neq 0 = \langle v_1, 0 \rangle = \langle v_1, Pv_3\rangle.$$

So, as you've observed, it's perfectly possible to take a linear operator that is not self-adjoint to a symmetric (or Hermitian) matrix, by way of a non-orthonormal basis. Just be aware of which theorems about inner product spaces require bases to be orthonormal to preserve inner product space properties.

  • $\begingroup$ Thank you for the clarification.. $\endgroup$ Mar 15 '21 at 8:06
  • $\begingroup$ Just to be clear, as we define these operators over $\mathbb{C}^n$, self adjoint and symmetric are equivalent right? $\endgroup$ Mar 15 '21 at 11:12
  • $\begingroup$ To be very clear, an operator $T$ on a finite-dimensional complex inner product space $V$ is self-adjoint if and only if $[T]_\beta^\beta$ is Hermitian (not symmetric!) where $\beta$ is an orthonormal basis for $V$. If $V = \Bbb{C}^n$, then the standard matrix for $V$ is the matrix for $T$ with respect to the standard basis, which is orthonormal, and hence $T$ is self-adjoint if and only if the standard matrix for $T$ is Hermitian. $\endgroup$ Mar 15 '21 at 11:17
  • $\begingroup$ Okay so orthonormal basis is the criteria here, any basis wont do $\endgroup$ Mar 15 '21 at 12:09
  • $\begingroup$ @roydiptajit Correct. $\endgroup$ Mar 15 '21 at 12:13

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.