I want to find the matrix representation of $S_z$ operator in terms of the eigenkets of $S_y$ operator.

( It's Problem 1.24 (b) of Sakurai Second Edition )

One simple way to solve this problem is to map (relabel) axes [x,y,z] to [z,x,y] ( note that the order of the axes is still a right-handed system ). With these new axes, $|$1 0 $\rangle$ and |0 1 $\rangle$ are eigenkets are $S_y$ operator. And so ( by comparison ), the matrix representation of operator $S_z$ is $\frac{\hbar}{2} \sigma_x$

Please correct me if the above solution is wrong. I verified this solution in an online solution manual of the textbook. ( This is not my main question though.)

However, I am trying to get the same solution by another approach, but I am getting a different ( wrong ) answer. The other approach is as follows:

Let $S_z$ be the abstract operator for measuring the z-component of spin ( whose matrix representation we want to find ). I can insert the identity operator to the right and the left side of $S_z$ to get:

$ S_z = \mathbb{1} S_z \mathbb{1} $

Now we can expand $\mathbb{1}$ in terms of the eigenkets of $S_y$ operator:

$ S_z = | S_y;+ \rangle \langle S_y;+ | S_z | S_y;+ \rangle \langle S_y;+ | \quad + $

$ \qquad | S_y;+ \rangle \langle S_y;+ | S_z | S_y;- \rangle \langle S_y;- | \quad + $

$ \qquad | S_y;- \rangle \langle S_y;- | S_z | S_y;+ \rangle \langle S_y;+ | \quad + $

$ \qquad | S_y;- \rangle \langle S_y;- | S_z | S_y;- \rangle \langle S_y;- | $

So, using $S_y$ basis, the $S_z$ operator can be written in the following matrix representation:

$$S_z^{(rep. \thinspace in \thinspace y \thinspace basis)} = \begin{bmatrix} \langle S_y;+ | S_z | S_y;+ \rangle & \langle S_y;+ | S_z | S_y;- \rangle \\ \langle S_y;- | S_z | S_y;+ \rangle & \langle S_y;- | S_z | S_y;- \rangle \end{bmatrix}$$

I just need to compute these elements of the matrix to get the required matrix representation.

Now I will use a concept here about which I am quite skeptical ( my main question is regarding the validity of this concept). I think that we can calculate these matrix elements in ANY BASIS ( not necessarily in $S_y$ basis ) because $\langle \alpha| O | \beta \rangle$ will give me the same value irrespective of which basis I choose to calculate it. After all, $ \alpha, \beta $ are abstract states and O is an abstract operator. So $\langle \alpha| O | \beta \rangle$ should give the same value is all basis/representations.

So, I proceed ahead and try to calculate these matrix elements in the standard $S_z$ basis ( ideally I should be calculating it in $S_y$ basis).

In standard z basis, we have :

$$S_z = \begin{bmatrix} \hbar/2 & 0 \\ 0 & -\hbar/2 \end{bmatrix}$$ $$| S_y;+ \rangle = \begin{bmatrix} 1/\sqrt{2} \\ i/\sqrt{2} \end{bmatrix}$$ $$| S_y;- \rangle = \begin{bmatrix} i/\sqrt{2} \\ 1/\sqrt{2} \end{bmatrix}$$

Using the above operators and states, I get this value for the representation of $S_z$ operator in $ | S_y ; \pm \rangle basis$ :

$$S_z^{(rep. \thinspace in \thinspace y \thinspace basis)} = -\frac{\hbar}{2} \sigma_y $$ which is $incorrect^*$.

I have spent good amount of time on this problem. Any help will be appreciated. To make my question concrete, here are the actual questions whose answers I want:

  1. Is my assumption ( given in boldface above ) correct?
  2. If it's incorrect, can you please tell me what's wrong with my understanding? Shouldn't the overlap of two states ( with a sandwiched operator ) be independent of basis?
  3. If you can find any other flaw in my understanding, that would also be very helpful.

[*]: As has been clarified in the answers below, this matrix representation is also correct. So the statement of my Stackexchange question is ( slightly ) wrong.


2 Answers 2


Your assumptions are not wrong, they are just incomplete. Let's start with the standard picture which is representing everything in the eigenbasis of $S_z$, meaning that $S_z = \hbar/2 \sigma^3$ using the Pauli matrices. The point here is that this is not enough to uniquely define $S_x$ and $S_y$. The reason is that any rotation about $S_z$ is also a valid choice for $S_x$ and $S_y$, while still respecting the right-handedness of the axes. That is, I can always define $S'_j = e^{i\theta S_z}S_j e^{-i\theta S_z}$ which will leave $S_z$ unchanged, but will change how the $x$ and $y$ axes are oriented. More explicitly, I can always do $$S_x = \frac{\hbar}{2} [\cos(\theta) \sigma^1 + \sin(\theta) \sigma^2]$$

$$S_y = \frac{\hbar}{2} [\cos(\theta) \sigma^2 - \sin(\theta) \sigma^1]$$ and everything remains ok.

This freedom of choice is encoded into the eigenstates by the arbitrary phase we can assign them. Indeed, a rotation about the $z$-axis will assign a phase to the eigenstates, which is physically insignificant.

If you want the "correct" result (the result you got is also correct, of course) you just have to choose $|S_y ; -\rangle = (|\!\uparrow\rangle - i |\!\downarrow\rangle)/\sqrt{2}$, which is multiplying your choice by $-i$.


Take, to avoid needless complication, $\hbar=2$, and observe the isomorphism of the algebra of the Pauli matrices, $$ \sigma_x \equiv \sigma_x ', \qquad \sigma_y \equiv -\sigma_z ', \qquad \sigma_z \equiv \sigma_y ', $$ That is, the σ' matrices obey the very same Lie algebra (commutators) as the σ ones, and we could use the standard representation of one set to (unconventionally) represent the other set.

So $\sigma_z$ in the basis of eigenvectors of $\sigma_y$ will be the same expression as $-\sigma_y$ in the basis of eigenvectors of $\sigma_z$.

Note that $\sigma_y \sigma_z=i \sigma_x=i \sigma_x'= \sigma_y' \sigma_z' $ . How did you conclude your fine answer is "incorrect"? You just rotated by -π/2 around the x-axis.

As @yu-v 's fine answer illustrates, any other algebra isomorphism will also do, like the cyclic permutation one you started with.


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.