Any Hamiltonian dynamic on a qubit is necessarily a rotation.

I consider a two level system having a Hamiltonian $H$. Decomposing this Hamiltonian on the Pauli basis and using the fact it is Hermitian, it is easy to show that it can be rewritten as: $$H=\frac{\hbar \Omega}{2}\overrightarrow{n}.\overrightarrow{\sigma}$$ This Hamiltonian implements the following unitary when evolved to a time $t$:

$$U_t = e^{-\frac{i \Omega t}{2} \overrightarrow{n}.\overrightarrow{\sigma}}$$

Such transformation induces a rotation in the Bloch-sphere around the axis $\overrightarrow{n}$ of an angle $\theta=\Omega t$.

Then, any (time-independent) Hamiltonian necessarily performs a rotation in the Bloch sphere. Why not.

Reflexion is a "valid" operation to do on a Bloch vector.

On the other hand, a qubit density matrix can be written as:

$$ \rho = \frac{1}{2} \left( \sigma_0 + \overrightarrow{p}.\overrightarrow{\sigma} \right)$$

Seeing it as a column vector $(1,\overrightarrow{p})$ (its decomposition in the Pauli basis), its evolution for a trace-preserving process can be computed through the Pauli-transfer matrix that has the following shape:

$$M=\begin{pmatrix} 1 & \overrightarrow{0} & \\ \overrightarrow{t} & A \end{pmatrix}$$

Where $A$ is a $3$x$3$ matrix, there are $0$'s on the first line for trace preservation. In the particular case $\overrightarrow{t}=0$ (unital map), $A$ is simply a matrix acting on $\overrightarrow{p}$. I could imagine $A$ being a reflexion: $A^T A= I$ but $det(A)=-1$

And this is where I get confused: We saw that a time-independent Hamiltonian necessarily induces a rotation. But as we can reason in $\mathbb{R}^3$ when seeing a qubit, reflections are perfectly valid operations to perform on it.

Does that imply that reflections are:

  • Not unitary operations when "translated" on the Bloch sphere
  • Are unitary operations, but cannot be implemented by a time-independent Hamiltonian (but I guess my example could be generalized as well to those cases...)

I would also be interested in a generalisation of this in higher dimension.

Note: I think this is very closely related to some notions of group theory, like the relationship between $U(n)$ and $O(n)$. But I have extremely bad memories of all that. So if an answer uses those notions, could you please remind me the very basic properties to know about this.


1 Answer 1


The reflection isn't unitary. The action of unitaries on the density matrix is by conjugation, which is trivial for the scalar matrices, giving a reduction

$$U(2) \to U(2)/U(1) = SO(3)$$

meaning there are only rotations, not reflections.

  • $\begingroup$ Thank you for your answer. Naive questions from a big beginner of group theory. $U(2)/U(1)$ is the quotient group. Physically here it means you "remove" the global phase from the description (you identify to a unique element all unitary differing from a global phase). Is that correct ? The equality with $SO(3)$ means isomorphic I guess. Do you have a reference for this ? $\endgroup$
    – StarBucK
    Commented Nov 25, 2020 at 0:35
  • $\begingroup$ Also. Is there an equivalent property in higher dimensions ? Up to a global phase $U(N)/U(1)$ is isomorphic to $SO(N+1)$ ? I am assuming I was correct in my previous comment. $\endgroup$
    – StarBucK
    Commented Nov 25, 2020 at 0:36
  • $\begingroup$ @StarBucK To your first comment: yes exactly. Two matrices in $U(2)$ which differ by $e^{i \theta}$ are identified. The isomorphism with $SO(3)$ is actually special to this case. For larger $N$ $U(N)/U(1)$, denoted $PU(N)$ for the projective unitary group, is just its own thing. To prove the isomorphism one can simply calculate that if we conjugate the matrix $\rho = 1 + a \sigma^x + b \sigma^y + c \sigma^z$ by $e^{i \theta \sigma_x}$, the effect is a rotation of the vector $(a,b,c)$ by an angle $2\theta$ about x. This gives us the map $U(2) \to SO(3)$, then we just need to find the kernel. $\endgroup$ Commented Nov 25, 2020 at 1:45
  • $\begingroup$ Thank you. I will try to see the proof on my side with your suggestion. So just to be sure: a $N>3$ dimension reflexion in $R^N$ could have a corresponding unitary matrix ? $\endgroup$
    – StarBucK
    Commented Nov 25, 2020 at 9:55
  • 1
    $\begingroup$ @StarBucK For general density matrices the action is given by the adjoint representation of $U(N)$ acting by conjugation on Hermitian matrices. This gives a map $U(N) \to SO(N^2-1)$. A simple reason it cannot contain a reflection is because $U(N)$ is connected, while $O(N^2-1)$ is disconnected. $\endgroup$ Commented Nov 25, 2020 at 22:32

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.