Ok hear me out.

One thing I have always liked about Dirac notation is that it visually displays where expressions expect inputs/outputs. For example $\langle\psi|$ expects an input to the right to form a complete expression like $\langle\psi|\phi\rangle$. The angle brackets indicate the expression is 'complete'. Things like $|\psi\rangle\langle\phi|$ require input both on the left and the right as indicated by the vertical bars. Now for tensor products this behaviour kind of breaks down. Expressions like $|n\rangle|\phi\rangle\equiv|n\rangle\otimes|\phi\rangle$ have two inputs but the input of the second ket is stuck behind the first ket. If you calculate something like $\langle m|\langle\psi|\langle\alpha||n\rangle|\phi\rangle|\beta\rangle=\langle m| n\rangle \langle\psi |\phi\rangle \langle\alpha |\beta\rangle$ the corresponding inputs and outputs are separated really far.

This might not sound like a big a deal (it isn't a big deal) but what if we stacked tensor products vertically? Like $$|n\rangle|\phi\rangle\equiv\begin{array}{c}|n\rangle\\|\phi\rangle\end{array}$$ Then the triple dot product from earlier becomes $$\begin{array}{c}\langle m|\\\langle\psi|\\\langle\alpha|\end{array}\cdot\begin{array}{c}|n\rangle\\|\phi\rangle\\|\beta\rangle\end{array}= \begin{array}{c}\langle m| n\rangle\\ \langle\psi |\phi\rangle \\\langle\alpha |\beta\rangle \end{array}= \langle m| n\rangle \langle\psi |\phi\rangle \langle\alpha |\beta\rangle$$ and products of general operators become $$(A\otimes B)\cdot(C\otimes D)=\begin{array}{c}A\\B\end{array}\cdot \begin{array}{c}C\\D\end{array}=\begin{array}{c}A\cdot C\\B\cdot D\end{array}=(A\cdot B)\otimes(C\cdot D)$$ So there are two benefits:

  1. Inputs/outputs are visually closer. So this is subjectively easier to understand.
  2. Expression with tensor products tend to work 'in parallel', that is each term in a tensor product often works separately from the others. The $|n\rangle$ and $|\phi\rangle$ in my example could be from entirely different Hilbert spaces. This vertical notation displays this visually because each tensor product term gets a separate row and the rows never talk to eachother.

Any thoughts? Improvements? I don't expect this notation will ever catch on because it is quite unwieldy but I hope this question gives you a fresh view on tensor products

  • 4
    $\begingroup$ I'm not sure this if this is on topic by my two cents are that I like that this mimics the way you'd lay out multiple registers in a quantum circuit (corresponding to multiple qubits) but dislike that it seems like it will take up lots of space. $\endgroup$
    – jacob1729
    Commented Jun 17, 2020 at 22:24
  • $\begingroup$ I really like this question but I fear it might be off topic, because it is just asking for opinions. $\endgroup$
    – Javier
    Commented Jun 17, 2020 at 23:07
  • 5
    $\begingroup$ Your notation can quickly become misleading: You might well visualize the one vector over the other as a concatenation of two vectors into one vector of dimensionality being the sum of those of the two. But that is the direct sum $\oplus$, and not the direct product $\otimes$. Multiplying the direct product by a number amounts to multiplying either factor by said number, but I defy you to intuit that in your vertical notation. Sooner or later, somebody clueless will slip and break his neck... $\endgroup$ Commented Jun 17, 2020 at 23:11
  • $\begingroup$ @Javier Yeah I was afraid of that. I hope that having soft-question as a tag prevents this question from being closed. $\endgroup$ Commented Jun 17, 2020 at 23:31
  • $\begingroup$ There is no reason to stop in 2D, some operators are easier to understand in 3D. But it is a hassle to write in 3D so it usually stops at proof of concept... $\endgroup$
    – Emil
    Commented Jun 18, 2020 at 6:58

1 Answer 1


Well, "de gustibus et de coloribus non est disputandum"... As indicated in my comment above, there is a red alert weakness, and steering users away from the evident enticing misunderstanding is tough. But you never know.

Suppose the dimension of $|n\rangle$ is a, take it to be 3, and that of $|\phi\rangle$ is b, take it to be 4.

Then the dimension of the vector $|n\rangle|\phi\rangle$ is ab, so, 12. However, the proposed notation, $$|n\rangle|\phi\rangle\equiv\begin{array}{c}|n\rangle\\|\phi\rangle\end{array}$$ all but invites the reader to think of the r.h.s. as the pileup of the two vectors, i.e. an a+b dimensional vector, here 7-d, (so $|n\rangle \oplus|\phi\rangle$), and not a 12-d vector.

You might beat your readers to not misunderstand the obvious, but somebody always will. (I am reminded of the trick-question calculus wag, who'd use the $x(f)$, $~~dx/df$, $~~d^2x/df^2$,... notation in trivial questions and stump his graders.) After virtually a century of Clebsching, it is a tall order to rescind collective public expectations.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.