In Shankar QM it is stated that since the $\boldsymbol K$ operator is Hermitian, vectors, which are expanded in the $\boldsymbol X$ basis with components $f(x) = \langle x | f \rangle$, must have an expansion in the $\boldsymbol K$ basis (i.e. the eigenbasis of $\boldsymbol K$).

Why is this?

  • 1
    $\begingroup$ This has nothing to do with being Hermitian. You can express vectors and operators in whatever basis you want. The definition of a basis is that you can do this. I think you aren't quite asking what you mean to ask. $\endgroup$ – DanielSank Jan 24 '15 at 4:19
  • $\begingroup$ That's what I thought, but he seemed to go out of the way to make the distinction. It was in a comment which set up the Fourier Transform as the Transformation from $\boldsymbol X$ to $\boldsymbol K$ $\endgroup$ – kypalmer Jan 24 '15 at 4:26
  • $\begingroup$ @DanielSank well the basis does have to be complete to do that. I'm guessing the $\boldsymbol K$ basis is supposed to be the eigenbasis of the operator, so the question boils down to whether the eigenbasis of a Hermitian operator is complete. kypalmer can you confirm? (and edit your question accordingly) $\endgroup$ – David Z Jan 24 '15 at 4:27
  • $\begingroup$ @DavidZ: What's a "basis" if it's not complete? To me "basis" means "complete set of vectors". $\endgroup$ – DanielSank Jan 24 '15 at 4:29
  • 1
    $\begingroup$ Again, you're asking why a vector can be expanded in two different bases. The answer is because a basis is defined as a set of vectors in which you can expand any other vector. $\endgroup$ – DanielSank Jan 24 '15 at 4:42

Theorem. If $\hat{K}$ is Hermitian then the eigenvectors of $\hat{K}$ can be chosen so as to form an orthonormal basis $|\xi_i\rangle$.

(The "can be chosen" wording completely handles the cases of degenerate eigenvalues, and the fact that an eigenvector times a constant is still an eigenvector)

(the converse is not true. In fact replace the word "Hermitian" with "Normal" and the theorem still holds. See: the wikipedia page on the spectral theorem. Also, I'm mainly considering the finite dimensional case)

Theorem. If $\{|\xi_i\rangle\}$ forms an orthonormal basis, then $I=\sum |\xi_i\rangle \langle \xi_i|$, where $I$ is the identity operator.

So, if your vector is $|\varphi\rangle$, you can also write it as $|\varphi\rangle=I|\varphi\rangle=\sum |\xi_i\rangle \langle \xi_i|\varphi\rangle$

if you're in a basis given by $\{ |n\rangle \}$, then $\langle n|\varphi\rangle$ are your components and $I=\sum |\xi_i\rangle \langle \xi_i|=\sum |n\rangle \langle n|$, so:

$|\varphi\rangle=II|\varphi\rangle=\sum_{i,n}|\xi_i\rangle \langle \xi_i|n\rangle \langle n|\varphi\rangle$

which is a change of basis represented by matrix multiplication, with $\langle \xi_i|n\rangle$ being the matrix elements.

My point in all this is that those two theorems when combined with Dirac notation are very nice!


Consider two particularly interesting bases:

  1. The basis of delta functions $\{|x\rangle\}$ where $\langle x | y\rangle =\delta(x - y)$. The $|x\rangle$ vectors are vectors with specific position in space.
  2. The basis of exponentials $\{|\nu \rangle\}$ where $\langle x | \nu \rangle = e^{i 2 \pi \nu x}$. These have specific spatial frequency. Note that $\langle \nu | \mu \rangle = \delta(\nu - \mu)$.

Now consider the equation we had at the end of the previous section:

\begin{align} |f\rangle &= \int dx d\nu | \nu \rangle \langle \nu | x \rangle \langle x | f \rangle \\ \langle \nu' | f \rangle &= \int dx d\nu \langle \nu'| \nu \rangle e^{-i 2 \pi \nu x} f(x)\\ &= \int dx d\nu \delta(\nu - \nu') e^{-i 2 \pi \nu px} f(x) \\ \tilde{f}(\nu') &= \int dx \, e^{-i 2 \pi \nu' x} f(x) \, . \end{align} So yeah, the Fourier transform is a change of basis.

  • 1
    $\begingroup$ By the way, using this nice notation you can explain Green's functions and essentially every use of linear algebra in physics. $\endgroup$ – DanielSank Jan 24 '15 at 4:52
  • $\begingroup$ So, basically, as long as both $ | \xi_i \rangle$ and $ |n\rangle$ are complete bases, then they are equivalent in their ability to describe any vector. Is this right? Does any of this change for an infinite dimensional function space? $\endgroup$ – kypalmer Jan 24 '15 at 4:59
  • $\begingroup$ @kypalmer: Glad you asked. Check out my edit to the answer. The "reason" the exponentials $\exp{i 2\pi \nu x}$ form a complete basis is that they are the eigenvectors (eigenfunctions) of the Hermitian operator $-iD$ where $D$ is the derivative with respect to $x$! I have written a very extensive and pedagogical document on this topic. If you would like me to email you a copy you can find my email address on my profile page. Don't be shy! I wrote it to help people like you. $\endgroup$ – DanielSank Jan 24 '15 at 5:07
  • $\begingroup$ There it is! Good extension to infinite dim. $\endgroup$ – kypalmer Jan 24 '15 at 5:13

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.