Reading the Nielsen and Chuang, I saw that every unitary operator $U$ can be written as $e^{i\alpha} R_n(\theta)$ for some well chosen $n \in \mathbb{R}^3$ and $0 \leq \theta < 2\pi$.

I would like to know if there is a simple criterion to ensure that $\alpha=0$, without knowing the value of $n$ or $\theta$.

As a consequence of it, using the fact that $e^{i\alpha}$ "disappear" when $U$ is seen as a rotation of the Bloch Sphere, and the fact that the composition of two rotations on $\mathbb{R}^3$ is also a rotation, I will have the proof that the composition of two morphisms $R_{n_1}(\theta_1) R_{n_2}(\theta_2)$ is on the form $R_{n_3}(\theta_3)$.

This problem is thus linked to the fact that the Bloch Sphere identify the unitary transformations that are equal up to a global phase (if I am right : $U_1$ and $U_2$, seen as morphisms on the Bloch sphere, are equal iff there are equal up to a global phase).

Edit: I precise I am interested by the result even if the global phase can be removed in one-qubit gates. One of the reasons is because of the multi-qubits gates, such as for exemple the two-bits gate that is a controled one-qubit gate.


2 Answers 2


As to your first question, I believe global phases can always be ignored, and we generally only concern ourselves with relative phases. So we don't really require that $\alpha=0$, we just ignore it.

As to your second point, when I read

are equal iff there are equal up to a global phase

I understand it as saying that they can have different global phases (since they don't matter). So you can have $U_1 = e^{i\alpha}R_n(\theta)$ and $U_2 = e^{i\beta}R_n(\theta)$, with $\alpha \neq \beta$, and we would say $U_1=U_2$ up to a global phase.

  • $\begingroup$ Actually, I am interested for the strict equality. One of the reasons is because of the multi-qubit gates, as I precised it in my edit. But, more generally, I would also be happy to know if it is true, even if it isn't relevant for the quantum computation applications. $\endgroup$ Feb 6, 2023 at 21:59
  • $\begingroup$ Not sure I understand your edit; we can still ignore global phases for n-qubit operations. Check out my answer here quantumcomputing.stackexchange.com/a/30077/19585, where we have an explicit example of ignoring global phases in a two-qubit gate. $\endgroup$
    – Bebotron
    Feb 6, 2023 at 22:40
  • $\begingroup$ Thanks for the link! I think there is indeed some point where I have a difficulty. My interest to keep the global phase for a one-qubit gate $U$ is to deal with controlled bit gate because I think that the global phase of $U$ becomes a relative phase if I study the controlled U gate. More generally, I think there is probably examples of multi-qubit gates V that are constructed using U, in such a way that the global phase of U only becomes relative phases in V and thus, has an effect. $\endgroup$ Feb 7, 2023 at 8:29
  • $\begingroup$ I'd be interested to see specific examples of gates you're looking at where this may be the case. It should not affect a controlled-$U$ operation. Consider the controlled operation $C(U)|a\rangle|b\rangle \rightarrow |a\rangle U|b\rangle$. You can always take out an arbitrary factor from $U$ to make it $e^{-i\alpha}U$, which again just gives you the same state up to a global phase factor $ |a\rangle e^{-i\alpha}U|b\rangle$, which can be ignored. Chapter 2 of Nielsen and Chuang should help you out. $\endgroup$
    – Bebotron
    Feb 7, 2023 at 9:23
  • 1
    $\begingroup$ For example, if I consider $U$ acting on the second qubit, controlled by the first qubit, and if I apply this gate to a state $\alpha |0\rangle| \phi \rangle + \beta | 1\rangle | \psi\rangle$, I think the result will be $| u\rangle= \alpha |0\rangle |\phi\rangle + \beta |1\rangle U(|\psi\rangle)$. If I replace $U$ by $e^{i\theta}U$, the result will instead be $|v\rangle = \alpha | 0\rangle |\phi\rangle + e^{i\theta}\beta |1\rangle U(|\psi\rangle)$, which is a change of relative phase, and thus $|u\rangle$ can be different of $|v\rangle$. $\endgroup$ Feb 8, 2023 at 20:01

$$ \text{det}(U)=1 $$ The rotations $R_n(\theta)$ are in the special unitary group, meaning that they have determinant 1. Hence, if you want $\alpha=0$, that would be achieved by having $\text{det}(U)=1$.

Sufficiency: let's assume that $\text{det}(U)=1$. This means that its eigenvalues must be of the form $e^{\pm i\theta}$. Let $P$ be the projector onto one of these eigenvalues, so $$ U=e^{i\theta}P+e^{-i\theta}(I-P). $$ Since $P$ is a projector (Hermitian, rank 1), and the Pauli matrices are a basis, we can always write $$ P=\frac{1}{2}(I+\vec{n}\cdot\vec{\sigma}) $$ where $\vec{n}$ is real-valued and has length 1. Thus, $$ U=I\cos\theta+i\sin\theta \vec{n}\cdot\vec{\sigma}=e^{i\theta\vec{n}\cdot\vec{\sigma}}. $$

  • $\begingroup$ By homogenity of the determinant, $\mathrm{det}(e^{i\alpha} U) = e^{i\alpha d}$ if $U\in SU(d)$. Hence, we can choose $\alpha = 2\pi k /d$ ($k\in\mathbb{Z}$) to get $\mathrm{det}(e^{i\alpha} U)=1$, i.e. $\alpha = 0, \pi$ for $d=2$. This is because the center of $SU(2)$ is $\{\pm I\}$. $\endgroup$ Mar 9, 2023 at 8:03
  • $\begingroup$ But if $\alpha=\pi$, there is an equivalent way of writing it such that $\alpha=0$, right? $\endgroup$
    – DaftWullie
    Mar 9, 2023 at 8:08
  • $\begingroup$ Yes indeed, we have $ - R_n(\theta) = R_n(\theta-\pi)$. I was missing that. Anyway, I doubt that a similar statement is true for for $d > 2$. $\endgroup$ Mar 9, 2023 at 13:35
  • $\begingroup$ @MarkusHeinrich agreed. $\endgroup$
    – DaftWullie
    Mar 9, 2023 at 13:57

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.