Single-Qubit Rotation Gates

Every operation a quantum computer performs on a single qubit is a $2\times2$ unitary, and once the physically invisible global phase is stripped away it is an element of $SU(2)$ — the group of unit-determinant $2\times2$ unitaries. That group is generated by exactly three matrices, the Pauli operators $X,Y,Z$ , so every single-qubit gate is the exponential of a real combination of Paulis. The rotation gates $R_x(\theta),R_y(\theta),R_z(\theta)$ are the cleanest slice of this fact: each is the exponential of a single Pauli about a single Cartesian axis, and together with rotations about tilted axes they sweep out all of $SU(2)$ . This article is where the discrete Pauli gates of the previous nodes become a continuous, angle-tunable family, and where the Bloch sphere stops being a static picture and starts to spin. Two threads run through everything below. First, an algebraic gift: because $\hat n\cdot\vec\sigma$ squares to the identity, its exponential resums into the finite closed form $\cos(\theta/2)\,I-i\sin(\theta/2)\,(\hat n\cdot\vec\sigma)$ — the Pauli plays the imaginary unit, and Euler's formula does the rest. Second, a geometric one: conjugating a state by $R_{\hat n}(\theta)$ rotates its Bloch vector rigidly by the full angle $\theta$ about $\hat n$ , though the gate itself carries only the half-angle $\theta/2$ . That mismatch is the fingerprint of the two-to-one map $SU(2)\to SO(3)$ — the reason a qubit is a spinor, and the reason a $2\pi$ turn flips a state's sign while a $4\pi$ turn is needed to truly do nothing. From here the Pauli gates reappear as half-turns, the Hadamard and phase gates as short rotation products, and the Z–Y decomposition delivers single-qubit universality.

Key takeaways

A rotation gate is the exponential of a Pauli about an axis: $R_{\hat n}(\theta)=e^{-i\theta\,\hat n\cdot\vec\sigma/2}$ , a continuous, angle-tunable single-qubit unitary.
Because $(\hat n\cdot\vec\sigma)^2=I$ , the exponential collapses to the closed form $\cos(\theta/2)\,I-i\sin(\theta/2)\,(\hat n\cdot\vec\sigma)$ — no infinite series survives.
On the Bloch sphere the gate is an honest rigid rotation by angle $\theta$ about $\hat n$ ; the gate carries $\theta/2$ , the sphere turns $\theta$ , and that factor of two is the $SU(2)\to SO(3)$ double cover.
Special angles recover the discrete gates: $R_x(\pi)=-iX$ , $R_z(\pi/2)=e^{-i\pi/4}S$ , and $R_{\hat n}(2\pi)=-I$ (not $I$ ).
Any single-qubit unitary factors as $U=e^{i\alpha}R_z(\beta)R_y(\gamma)R_z(\delta)$ (Z–Y decomposition), so three rotations plus a phase realise all of $SU(2)$ — the root of single-qubit universality.
Rotations are the native, calibrated primitives of real hardware: virtual- $Z$ frame changes, resonant $R_x(\pi/2)$ pulses, and the parameter-shift gradients that train variational circuits.
Key idea: each Pauli squares to identity, so its exponential resums to $R_{\hat n}(\theta)=\cos(\theta/2)\,I-i\sin(\theta/2)\,(\hat n\cdot\vec\sigma)$ — a half-angle $SU(2)$ gate that drives a full-angle $SO(3)$ rotation of the Bloch vector.

The core insight

Each Pauli squares to identity, so its exponential resums into a finite Euler formula, $R_{\hat n}(\theta)=\cos(\theta/2)\,I-i\sin(\theta/2)\,(\hat n\cdot\vec\sigma)$ — a half-angle rotation in $SU(2)$ that rotates the Bloch vector by the full angle $\theta$ about $\hat n$ in $SO(3)$ .

Where this lives today

Rotation gates are not a pedagogical abstraction; they are the literal instruction set of every quantum processor now running. On superconducting hardware the $Z$ -rotation is a virtual- $Z$ gate implemented for free — the control software advances the phase reference of all subsequent microwave pulses, so $R_z(\theta)$ takes zero time and zero error (McKay and co-workers, 2017); the only driven single-qubit gate is then a calibrated $R_x(\pi/2)$ (the " $\sqrt{X}$ "), from which any rotation compiles as $R_z R_x R_z$ . Trapped-ion machines expose driven $R_x(\theta),R_y(\theta)$ directly, the axis set by the drive phase and the angle by the pulse area $\Omega t$ . Above the hardware, variational algorithms (VQE, QAOA) build ansätze from parametrised rotations and train them by the parameter-shift rule, an exact gradient that exploits the generator's two-eigenvalue structure. And at the fault-tolerant frontier the Solovay–Kitaev theorem and number-theoretic compilers (Ross–Selinger) approximate any $R_z(\theta)$ to accuracy $\varepsilon$ with $O(\log(1/\varepsilon))$ Clifford+ $T$ gates — the bridge from continuous angles to the discrete, error-corrected circuits a scalable machine will run.

The physical picture

Hold the operator $\hat n\cdot\vec\sigma$ in your hand as if it were the number $i$ . It is Hermitian, traceless, and — the crucial fact — it squares to the identity, exactly the property $i^2=-1$ upgraded to a matrix. Feed it into an exponential and you do not get an unruly infinite series; you get sines and cosines, because the even powers pile up on $I$ and the odd powers pile up on $\hat n\cdot\vec\sigma$ , just as $e^{i\phi}$ splits into $\cos\phi$ and $i\sin\phi$ . A rotation gate is therefore a smooth, sinusoidal dial between doing nothing (the $I$ term at $\theta=0$ ) and applying the full Pauli reflection (the $\hat n\cdot\vec\sigma$ term at $\theta=\pi$ ). Geometrically that dial is a turntable for the Bloch sphere: choose an axis, choose an angle, and spin the whole sphere. The first figure makes the half-angle-in, full-angle-out relationship concrete for one specific gate.

Figure

Bloch-sphere turntable: the gate R_x(π/2) = cos(π/4)I − i sin(π/4)X = 0.7071(I − iX) rotates the state's Bloch vector rigidly by the full angle θ = 90° about x̂, carrying the north pole (the 0 state) to the −ŷ point, while the gate itself carries only the half-angle θ/2 = 45° — the fingerprint of the SU(2)→SO(3) double cover

Push the dial further and the double cover announces itself. As $\theta$ runs through $\pi$ , $2\pi$ , and beyond, the matrix entries $\cos(\theta/2)$ and $\sin(\theta/2)$ trace sinusoids of period $4\pi$ , not $2\pi$ . A full $2\pi$ turn brings the Bloch vector exactly home, yet the state vector has quietly picked up a global $-1$ : the gate is $-I$ , not $I$ . Only after a second full turn, at $\theta=4\pi$ , does it return to the true identity. This holds for every axis $\hat n$ , and it is the same $-1$ a neutron beam accumulates when its spin is precessed through $2\pi$ . The second figure plots both entries across two full Bloch turns so the $4\pi$ periodicity is impossible to miss.

Figure

The double cover made visible: the matrix entries of R_n(θ) are cos(θ/2) and sin(θ/2), so the gate has period 4π rather than 2π — at θ = 2π the gate equals −I, a full Bloch turn that flips the state's global sign, and only at θ = 4π does it return to the identity I

The core equations

The defining object is the exponential of a Pauli about an axis. A real unit vector $\hat n$ picks the axis, the angle $\theta$ sets the amount of turn, and the generator is $-i/2$ times the corresponding component of the Pauli vector — the $-i$ makes the result unitary, and the $\tfrac12$ is the half-angle that the double cover forces on us.

R_{\hat n}(\theta)=\exp\!\left(-i\,\frac{\theta}{2}\,\hat n\cdot\vec\sigma\right) \qquad (1)

where: $\theta$ — rotation angle on the Bloch sphere (radians); $\hat n=(n_x,n_y,n_z)$ — real unit rotation axis; $\vec\sigma=(X,Y,Z)$ — the Pauli vector; $\hat n\cdot\vec\sigma=n_xX+n_yY+n_zZ$ — a traceless Hermitian involution. Reading: the generator $\hat n\cdot\vec\sigma$ lives in the Lie algebra $\mathfrak{su}(2)$ ; exponentiating it sweeps out the group $SU(2)$ . The single most useful identity in the subject is that this exponential is finite. Because the generator squares to the identity, the Taylor series folds up exactly into a cosine on $I$ and a sine on the generator — Euler's formula with a matrix in the seat of $i$ .

R_{\hat n}(\theta)=\cos\!\frac{\theta}{2}\,I-i\sin\!\frac{\theta}{2}\,(\hat n\cdot\vec\sigma) \qquad (2)

where: $I$ — the $2\times2$ identity. Reading: every single-qubit rotation is one cosine, one sine, and the axis — nothing else to compute, and the derivation below shows why it ends. Specialising the axis to $\hat x$ , $\hat y$ , $\hat z$ gives the three canonical rotation matrices. Written out, $R_x$ and $R_z$ are complex while $R_y$ is the lone all-real member — a fact that makes $R_y$ the workhorse of amplitude-only rotations.

R_x(\theta)=\begin{pmatrix}\cos\frac{\theta}{2}&-i\sin\frac{\theta}{2}\\[2pt]-i\sin\frac{\theta}{2}&\cos\frac{\theta}{2}\end{pmatrix} \qquad (3)

R_y(\theta)=\begin{pmatrix}\cos\frac{\theta}{2}&-\sin\frac{\theta}{2}\\[2pt]\sin\frac{\theta}{2}&\cos\frac{\theta}{2}\end{pmatrix} \qquad (4)

R_z(\theta)=\begin{pmatrix}e^{-i\theta/2}&0\\0&e^{i\theta/2}\end{pmatrix}=e^{-i\theta/2}\,\mathrm{diag}\!\left(1,e^{i\theta}\right) \qquad (5)

where: each $\theta$ is the turn about the named axis. Reading: $R_z$ is diagonal because $Z$ is; the rightmost form of (5) exposes that $R_z(\theta)$ equals the phase gate $\mathrm{diag}(1,e^{i\theta})$ only up to the global factor $e^{-i\theta/2}$ — a distinction the Common-pitfall box below insists on. The geometric payoff is that conjugating the Bloch-vector operator by a rotation gate produces a genuine $3\times3$ real rotation of the Bloch vector. The gate acts on $2$ -component amplitudes; its shadow on $\mathbb R^3$ is an ordinary rotation matrix.

R_{\hat n}(\theta)\,(\vec r\cdot\vec\sigma)\,R_{\hat n}(\theta)^\dagger=\big(\mathcal R_{\hat n}(\theta)\,\vec r\big)\cdot\vec\sigma \qquad (6)

where: $\vec r$ — Bloch vector, $\rho=\tfrac12(I+\vec r\cdot\vec\sigma)$ ; $\mathcal R_{\hat n}(\theta)\in SO(3)$ — the $3\times3$ rotation by angle $\theta$ about $\hat n$ . Reading: the map $R_{\hat n}(\theta)\mapsto\mathcal R_{\hat n}(\theta)$ is the two-to-one homomorphism $SU(2)\to SO(3)$ ; both $\pm R_{\hat n}(\theta)$ map to the same $\mathcal R_{\hat n}(\theta)$ . Two rotations about different axes compose into a single rotation about a third — the quaternion multiplication rule in disguise, and the algebraic core of gate synthesis.

R_{\hat a}(\theta)R_{\hat b}(\phi)=\big[c_\theta c_\phi-s_\theta s_\phi\,(\hat a\!\cdot\!\hat b)\big]I-i\big[c_\theta s_\phi\,\hat b+s_\theta c_\phi\,\hat a+s_\theta s_\phi\,(\hat a\!\times\!\hat b)\big]\!\cdot\!\vec\sigma \qquad (7)

where: $c_\theta=\cos(\theta/2),\ s_\theta=\sin(\theta/2)$ (likewise for $\phi$ ); $\hat a,\hat b$ — the two axes. Reading: the scalar part gives the resultant half-angle via $\cos(\Theta/2)=c_\theta c_\phi-s_\theta s_\phi(\hat a\cdot\hat b)$ ; the vector part gives the resultant axis. Finally, three rotations exhaust the group: any single-qubit unitary is a global phase times a $Z$ – $Y$ – $Z$ sandwich. This is the Z–Y decomposition, and it is the reason the rotation family alone is universal for one qubit.

U=e^{i\alpha}\,R_z(\beta)\,R_y(\gamma)\,R_z(\delta) \qquad (8)

where: $\alpha$ — global phase in $U(1)$ ; $\beta,\gamma,\delta$ — three real Euler angles fixing the $SU(2)$ part. Reading: $U(2)$ has four real parameters; one is the phase $\alpha$ and three are the Euler angles, matching the count exactly.

Derivation AResumming the exponential into a closed-form rotation

The goal is to prove that the matrix exponential in equation (1) is not an infinite object at all but the two-term closed form of equation (2), and to see precisely which property of the generator makes the collapse happen. The lever is a single algebraic fact — that $\hat n\cdot\vec\sigma$ is an involution, squaring to the identity — which lets us sort the power series into even and odd terms and recognise the cosine and sine series hiding inside. This is the exact analogue of deriving Euler's formula $e^{i\phi}=\cos\phi+i\sin\phi$ from $i^2=-1$ , with the Pauli combination playing the role of $i$ .

Step 1. Abbreviate $A=\hat n\cdot\vec\sigma$ and expand the exponential as a power series, $e^{-i(\theta/2)A}=\sum_{k\ge0}\frac{1}{k!}\left(-i\tfrac{\theta}{2}\right)^k A^k$ . Why: nothing is assumed yet — this is just the definition of a matrix exponential, valid because the series converges absolutely for any finite matrix.

Step 2. Show $A^2=I$ . Squaring, $A^2=\sum_{j,k}n_jn_k\,\sigma_j\sigma_k$ ; the symmetric part uses $\sigma_j\sigma_j=I$ and the antisymmetric part cancels because distinct Paulis anticommute, $\sigma_j\sigma_k+\sigma_k\sigma_j=0$ . What survives is $(n_x^2+n_y^2+n_z^2)I=I$ since $\hat n$ is a unit vector. Why: this is the one fact that makes the whole series terminate — every power of $A$ is now either $I$ or $A$ .

Step 3. Split by parity. Even powers give $A^{2m}=I$ and odd powers give $A^{2m+1}=A$ , so the sum separates into $\big(\sum_m\frac{(-1)^m(\theta/2)^{2m}}{(2m)!}\big)I-i\big(\sum_m\frac{(-1)^m(\theta/2)^{2m+1}}{(2m+1)!}\big)A$ . Why: gathering the $I$ terms and the $A$ terms is legitimate because they are linearly independent, so no cross-contamination occurs.

Step 4. Recognise the two bracketed series as $\cos(\theta/2)$ and $\sin(\theta/2)$ , giving $R_{\hat n}(\theta)=\cos(\theta/2)\,I-i\sin(\theta/2)\,A$ . Why: this is equation (2); the exponential of any axis-Pauli is a finite, hand-computable $2\times2$ matrix — the practical foundation for everything that follows. *As a concrete check: $R_x(\pi/2)=0.7071(I-iX)$ and $R_z(\pi)=-iZ$ reproduce the closed form exactly, and the involution $A^2=I$ holds for a tilted axis such as $\hat n=(1,1,-1)/\sqrt3$ .*

Go deeper · Derivation B · 1 minConjugation turns the gate into a rigid Bloch-sphere rotation

Derivation BConjugation turns the gate into a rigid Bloch-sphere rotation

The goal is to demonstrate equation (6) concretely for $R_z(\theta)$ and, in doing so, to expose the factor of two between the gate's argument and the geometric turn. We compute how $R_z(\theta)$ conjugates the operator $X$ ; the result is a linear combination of $X$ and $Y$ with coefficients $\cos\theta$ and $\sin\theta$ — the full angle, not the half-angle — which is exactly a rotation of the $x$ – $y$ plane of the Bloch sphere about the $z$ -axis. This is where the half-angle in the gate and the full angle on the sphere are reconciled, and it corrects a sign that is easy to drop.

Step 1. Write the conjugation in generator form, $R_z(\theta)XR_z(\theta)^\dagger=e^{-i\theta Z/2}\,X\,e^{i\theta Z/2}$ , and apply the Hadamard lemma $e^{G}Xe^{-G}=X+[G,X]+\tfrac12[G,[G,X]]+\cdots$ with $G=-i\theta Z/2$ . Why: the nested-commutator expansion turns operator conjugation into a series in $\theta$ that we can resum, mirroring Step 1 of the previous derivation.

Step 2. Evaluate the first commutator using $[Z,X]=2iY$ , so $[G,X]=-\tfrac{i\theta}{2}(2iY)=+\theta\,Y$ . The double commutator uses $[Z,Y]=-2iX$ , giving $[G,[G,X]]=-\theta^2X$ , and the pattern then cycles $X\to Y\to -X\to -Y$ . Why: the commutators reproduce the derivative structure of sine and cosine, which is the algebraic reason a rotation appears.

Step 3. Resum the two interleaved series to $R_z(\theta)XR_z(\theta)^\dagger=\cos\theta\,X+\sin\theta\,Y$ . A direct matrix multiplication confirms the sign: $R_z(\theta)XR_z(\theta)^\dagger$ has off-diagonal entries $e^{\mp i\theta}$ , which equal $\cos\theta\,X+\sin\theta\,Y$ , not the minus-sign variant. Why: keeping this sign correct is what makes the operator identity consistent with the state-vector rotation in the next step — a subtlety the thin migrated draft got backwards.

Step 4. Translate to the Bloch vector. Since $r_i'=\mathrm{Tr}(\rho\,R^\dagger\sigma_iR)$ , the transformed components are $r_x'=r_x\cos\theta-r_y\sin\theta$ , $r_y'=r_x\sin\theta+r_y\cos\theta$ , $r_z'=r_z$ — the standard $SO(3)$ rotation by the full angle $\theta$ about $\hat z$ . Why: the gate argument is $\theta/2$ but the sphere turns by $\theta$ , precisely the two-to-one $SU(2)\to SO(3)$ relationship, verified against the state $R_z(\theta)|+\rangle$ whose Bloch vector is $(\cos\theta,\sin\theta,0)$ . *As a concrete check: $R_z(0.7)XR_z(0.7)^\dagger=\cos(0.7)X+\sin(0.7)Y$ ; the alternative $-\sin\theta\,Y$ form gives the wrong sign and fails this check.*

Go deeper · Derivation C · 1 minEvery single-qubit gate is three rotations

Derivation CEvery single-qubit gate is three rotations

The goal is to prove the Z–Y decomposition of equation (8): that an arbitrary $U\in U(2)$ can be written as a global phase times $R_z(\beta)R_y(\gamma)R_z(\delta)$ , so the rotation gates alone generate every single-qubit operation. The strategy is to strip the global phase to reduce to $SU(2)$ , note that any $SU(2)$ matrix has a rigid two-parameter shape, and then show that the explicit product $R_z(\beta)R_y(\gamma)R_z(\delta)$ realises that shape for suitable angles. This is the concrete engine behind single-qubit universality and behind every compiler that turns an abstract gate into hardware rotations.

Step 1. Factor the phase. Any $U\in U(2)$ has $|\det U|=1$ , so $\det U=e^{2i\alpha}$ for some real $\alpha$ ; then $e^{-i\alpha}U\in SU(2)$ . Every special-unitary $2\times2$ matrix has the form $\begin{pmatrix}a&-\bar b\\ b&\bar a\end{pmatrix}$ with $|a|^2+|b|^2=1$ . Why: this two-complex-parameter shape, constrained by one real equation, carries exactly the three real degrees of freedom of $SU(2)$ .

Step 2. Multiply out the ansatz. Using equations (3)–(5), $R_z(\beta)R_y(\gamma)R_z(\delta)$ has top-left entry $e^{-i(\beta+\delta)/2}\cos(\gamma/2)$ and bottom-left entry $e^{i(\beta-\delta)/2}\sin(\gamma/2)$ , and the other two entries are fixed by the $\begin{pmatrix}a&-\bar b\\ b&\bar a\end{pmatrix}$ pattern. Why: the product already lands in $SU(2)$ , so we only need to hit the correct $a$ and $b$ .

Step 3. Match parameters. Set $a=e^{-i(\beta+\delta)/2}\cos(\gamma/2)$ and $b=e^{i(\beta-\delta)/2}\sin(\gamma/2)$ ; then $|a|=|\cos(\gamma/2)|$ and $|b|=|\sin(\gamma/2)|$ fix $\gamma$ , while the two phases fix $\beta+\delta$ and $\beta-\delta$ , hence $\beta$ and $\delta$ separately. Why: three real equations for three angles always have a solution, so no $SU(2)$ target is unreachable.

Step 4. Reassemble. Restoring the phase gives $U=e^{i\alpha}R_z(\beta)R_y(\gamma)R_z(\delta)$ , the claimed decomposition. Why: the same argument with $R_y$ replaced by $R_x$ , or $R_z$ by $R_x$ , gives the Z–X and X–Y variants — the freedom compilers exploit to match a device's native axes. *As a concrete check: for $(\beta,\gamma,\delta)=(0.9,1.3,0.4)$ the assembled product's entries match $a$ and $b$ exactly.*

Limiting cases

Sending the angle or axis to special values recovers the discrete gates and the double-cover signature — each an entry the rotation family "contains" up to a global phase.

Case	Result	Meaning
$\theta=\pi$ about $\hat x$	$R_x(\pi)=-iX$	a half-turn about $\hat x$ is the Pauli $X$ gate up to the unobservable phase $-i$
$\theta=2\pi$ , any axis	$R_{\hat n}(2\pi)=-I$	a full Bloch turn returns the vector but flips the state's sign — the spinor double cover
$\theta=\pi/2$ about $\hat z$	$R_z(\pi/2)=e^{-i\pi/4}S$	recovers the phase gate $S$ up to a phase; likewise $R_z(\pi/4)=e^{-i\pi/8}T$
$\theta\to0$	$R_{\hat n}(\theta)\approx I-\tfrac{i\theta}{2}\,\hat n\cdot\vec\sigma$	infinitesimal rotations are generated by the Pauli; the generator is $-\tfrac{i}{2}\hat n\cdot\vec\sigma$

Common trapThe gate carries θ/2 but the sphere turns θ — and R_z(θ) is not diag(1, e^{iθ})

Two linked traps. First, the factor of two: the exponential's argument is $\theta/2$ , yet the Bloch vector rotates by the full $\theta$ . A circuit's $R_x(\pi)$ is a $180^\circ$ Bloch turn (a bit flip), not a $90^\circ$ one; a genuine $90^\circ$ turn needs $R_x(\pi/2)$ . Second, and more insidious, $R_z(\theta)$ is not the phase gate $P(\theta)=\mathrm{diag}(1,e^{i\theta})$ ; they differ by a global phase, $R_z(\theta)=e^{-i\theta/2}P(\theta)=\mathrm{diag}(e^{-i\theta/2},e^{i\theta/2})$ . Acting alone the two are physically identical, so you may use either. But inside a controlled gate that "global" phase becomes a relative phase between the control branches and is fully observable: controlled- $R_z(\theta)$ and controlled- $P(\theta)$ are different unitaries. Compilers and textbooks split on which they call " $R_z$ ", so always check whether a diagonal phase or a symmetric $\pm\theta/2$ phase is meant.

Worked examples

1 · Composing two π/2 rotations into a single axis-angle turn (fully worked)

A circuit applies $R_x(\pi/2)$ and then $R_y(\pi/2)$ to a qubit. Every product of rotations is again a rotation — find the single axis $\hat n$ and angle $\Theta$ of the combined gate $U=R_y(\pi/2)R_x(\pi/2)$ .

Step 1. Write each factor in closed form. With $\cos(\pi/4)=\sin(\pi/4)=1/\sqrt2$ , both are $R_x(\pi/2)=\tfrac{1}{\sqrt2}(I-iX)$ and $R_y(\pi/2)=\tfrac{1}{\sqrt2}(I-iY)$ .

Step 2. Multiply. $U=\tfrac12(I-iY)(I-iX)=\tfrac12\big(I-iX-iY+i^2\,YX\big)$ . Using $YX=-iZ$ , the last term is $i^2(-iZ)=+iZ$ , so $U=\tfrac12 I-\tfrac{i}{2}(X+Y-Z)$ .

Step 3. Read off the axis-angle by matching to $\cos(\Theta/2)I-i\sin(\Theta/2)(\hat n\cdot\vec\sigma)$ . The identity coefficient gives $\cos(\Theta/2)=\tfrac12$ , so $\Theta/2=60^\circ$ and $\boldsymbol{\Theta=120^\circ}$ . The Pauli coefficient is $\tfrac12(X+Y-Z)$ ; dividing by $\sin(60^\circ)=\sqrt3/2$ gives $\hat n\cdot\vec\sigma=\tfrac{1}{\sqrt3}(X+Y-Z)$ , hence $\boldsymbol{\hat n=(1,1,-1)/\sqrt3}$ .

Step 4. Cross-check with the scalar composition law (7): $\cos(\Theta/2)=c_\theta c_\phi-s_\theta s_\phi(\hat x\cdot\hat y)=\tfrac{1}{\sqrt2}\cdot\tfrac{1}{\sqrt2}-\tfrac12\cdot0=\tfrac12$ , agreeing with $\Theta=120^\circ$ ; and $\hat n$ is a unit vector since $(1+1+1)/3=1$ . Two orthogonal $90^\circ$ turns compose into one $120^\circ$ turn about the body-diagonal $(1,1,-1)/\sqrt3$ — a fact any circuit optimiser uses to fuse adjacent rotations.

2 · Preparing an equal superposition with a π/2 pulse

A qubit starts in $|0\rangle$ . Apply $R_x(\pi/2)$ . Find the output state, the measurement probabilities in the computational basis, and the Bloch vector.

Step 1. Apply the gate. $R_x(\pi/2)|0\rangle=\tfrac{1}{\sqrt2}(I-iX)|0\rangle=\tfrac{1}{\sqrt2}\big(|0\rangle-i|1\rangle\big)$ .

Step 2. Read the probabilities. $P(0)=|\langle0|\psi\rangle|^2=\tfrac12$ and $P(1)=|{-i}/\sqrt2|^2=\tfrac12$ — an equal superposition, $P_0=P_1=\tfrac12$ , just like a Hadamard but with a relative phase of $-i$ instead of $+1$ .

Step 3. Find the Bloch vector. A state $\tfrac{1}{\sqrt2}(|0\rangle+e^{i\varphi}|1\rangle)$ has Bloch vector $(\cos\varphi,\sin\varphi,0)$ ; here $e^{i\varphi}=-i$ so $\varphi=-\tfrac{\pi}{2}$ and $\vec r=(0,-1,0)=\boldsymbol{-\hat y}$ . The north pole $|0\rangle$ has rotated $90^\circ$ about $\hat x$ down to the $-\hat y$ point — exactly the trajectory drawn in the first figure, and the physical content of a $\pi/2$ drive pulse in NMR or a superconducting qubit.

Physics of the portrait

The rotation gate is a piece of nineteenth-century mathematics wearing a twentieth-century physical costume. Hamilton's quaternions already encoded the double cover of rotations in 1843; what was missing was a physical system whose states actually transform under $SU(2)$ rather than $SO(3)$ . That system arrived with the electron's spin, and the person who wrote down the generators we now exponentiate — the matrices $X,Y,Z$ — was Wolfgang Pauli.

Origins · 1 minWolfgang Pauli and how this was discovered

Wolfgang Pauli

1900–1958 · the spin matrices and the algebra behind single-qubit rotations

Wolfgang Pauli was a physicist of ferocious rigour. His decisive contribution here is the 1927 paper Zur Quantenmechanik des magnetischen Elektrons, where, to describe the electron's newly discovered two-valued spin, he introduced the three $2\times2$ matrices $\sigma_x,\sigma_y,\sigma_z$ that now bear his name. These are precisely the generators whose exponentials are the rotation gates, $R_{\hat n}(\theta)=\exp(-i\theta\,\hat n\cdot\vec\sigma/2)$ . Pauli was not thinking of computing; he was capturing how a spin- $\tfrac12$ magnetic moment responds to a field, and the half-angle that puzzles students of gates was for him the plain statement that a spin lives in the double cover of the rotation group — a $2\pi$ turn returns the direction but negates the wavefunction. His 1925 exclusion principle had already forced the electron to carry a hidden two-valued label; the spin matrices gave that label its algebra. A century on, every calibrated $R_x(\pi/2)$ pulse on a real qubit is an experimental realisation of Pauli's matrices, and every compiler that decomposes a gate into $R_z R_y R_z$ is doing group theory Pauli would have recognised at once.

History

1843

— Hamilton discovers the quaternions, the algebra of SU(2) rotations

1925

— Pauli's exclusion principle forces a hidden two-valued electron degree of freedom

1927

— Pauli introduces the 2×2 spin matrices X, Y, Z that generate the rotation gates

1931

— Wigner brings SU(2) and its two-to-one cover of SO(3) into quantum theory

1938

— Rabi drives resonant spin flips, realizing R_n(θ) as a timed pulse

1995

— Barenco and co-workers prove single-qubit rotations plus CNOT are universal

Going further

Texts. Nielsen & Chuang, Quantum Computation and Quantum Information, §4.2 (single-qubit operations) and §4.3 · Preskill, Lecture Notes on Quantum Computation, Ch. 2–3 · Watrous, The Theory of Quantum Information, Ch. 2 · Kitaev, Shen & Vyalyi, Classical and Quantum Computation, §8 · Mermin, Quantum Computer Science, Ch. 1 · Barenco et al., "Elementary gates for quantum computation", Phys. Rev. A 52, 3457 (1995). Key results to keep in mind: (1) the resummation from $A^2=I$ yields the closed form; (2) the explicit matrices satisfy $R_z(\theta)=e^{-i\theta/2}\mathrm{diag}(1,e^{i\theta})$ ; (3) $R_{\hat n}(\pi)=-i\hat n\cdot\vec\sigma$ , recovering the Pauli gates; (4) the conjugation $R_z(\theta)XR_z(\theta)^\dagger=\cos\theta\,X+\sin\theta\,Y$ via the Hadamard lemma with $[Z,X]=2iY$ — note the $+$ sign, confirmed by direct matrix multiplication; (5) the composition $R_y(\pi/2)R_x(\pi/2)=\tfrac12 I-\tfrac{i}{2}(X+Y-Z)$ , a $120^\circ$ turn about $(1,1,-1)/\sqrt3$ ; (6) the Z–Y entries are $a=e^{-i(\beta+\delta)/2}\cos(\gamma/2)$ and $b=e^{i(\beta-\delta)/2}\sin(\gamma/2)$ .

Builds on. Pauli Gates (X, Y, Z) · Bloch Sphere Representation of Gates · Quantum Circuit Model. Leads to. Z–Y Decomposition of Single-Qubit Gates · Phase Gates (S and T) · Controlled-U Gates · Universal Gate Sets.

Ask the tutor: Ask me to walk you through the Z–Y decomposition of any specific gate you name, or to show how a resonant pulse of a given area realises it on real hardware.

Why does the gate carry half the angle that the Bloch sphere rotates through?

Because a qubit is a spinor, not a direction in space. Its state lives in a complex $2$ -dimensional space on which $SU(2)$ acts, and $SU(2)$ is the double cover of $SO(3)$ : the map from gates to physical rotations is two-to-one. The gate $R_{\hat n}(\theta)$ and its negative $-R_{\hat n}(\theta)$ give the identical $SO(3)$ rotation of the Bloch vector, so a single geometric turn corresponds to two gate values differing by $\theta\to\theta+2\pi$ . The half-angle in the exponent is exactly what makes one full geometric turn ( $\theta=2\pi$ ) send the gate to $-I$ rather than back to $I$ — you are seeing the spinor nature of the qubit directly in the arithmetic.

Is R_z(θ) the same as the phase gate diag(1, e^{iθ})?

Not as matrices, only as operations on a lone qubit. Writing them out, $R_z(\theta)=\mathrm{diag}(e^{-i\theta/2},e^{i\theta/2})$ while $P(\theta)=\mathrm{diag}(1,e^{i\theta})$ ; they differ by the overall factor $e^{-i\theta/2}$ . A global phase is unobservable, so acting alone the two give identical statistics and Bloch rotations, and you may swap them freely. The catch is control: in a controlled version the phase that was global for the target becomes a relative phase between the "control off" and "control on" branches, now fully measurable. So controlled- $R_z(\theta)$ and controlled- $P(\theta)$ are genuinely different gates — always check which convention a paper or compiler means by " $R_z$ ".

How do I get an arbitrary single-qubit gate if hardware only gives me rotations?

You use the Z–Y decomposition: every $U\in U(2)$ equals $e^{i\alpha}R_z(\beta)R_y(\gamma)R_z(\delta)$ for some phase and three Euler angles. Finding them is short — read $\gamma$ off the entry magnitudes and $\beta,\delta$ off their phases. If your device's native axes are not $y$ and $z$ , an equivalent Z–X or X–Y form does the same job, since the derivation needs only two non-parallel axes. This is exactly what a compiler does: it takes the abstract unitary and emits a short list of native rotations — which is why "single-qubit universality" and "three rotations" are the same statement.

What is the physical meaning of R_n(θ) on real qubit hardware?

It is a driven oscillation. Apply a resonant pulse — microwave for superconducting qubits, laser or microwave for ions — of fixed amplitude for a time $t$ , and the qubit undergoes $R_{\hat n}(\Omega t)$ , where $\Omega$ is the Rabi frequency and the equatorial axis $\hat n$ is set by the phase of the drive. A pulse area $\Omega t=\pi$ gives a $\pi$ -pulse, a bit flip; $\Omega t=\pi/2$ gives the $\pi/2$ -pulse that creates superpositions and opens and closes Ramsey interferometry. A rotation gate is thus no metaphor: it is the closed-form solution of the driven two-level Schrödinger equation, and the angle is literally how long you drive.

Why is R_y the only rotation with real matrix entries, and does that matter?

It comes down to which Pauli is imaginary. Of $X,Y,Z$ , only $Y=\begin{pmatrix}0&-i\\ i&0\end{pmatrix}$ is itself imaginary, so $-i\sin(\theta/2)\,Y$ comes out real, leaving $R_y(\theta)$ with entirely real entries $\cos(\theta/2)$ and $\pm\sin(\theta/2)$ ; $R_x$ has imaginary off-diagonals and $R_z$ complex diagonal phases. This is more than a curiosity: because $R_y$ keeps amplitudes real, it is the natural gate for adjusting the populations of $|0\rangle$ and $|1\rangle$ without adding relative phase — which is why the Z–Y decomposition puts $R_y$ in the middle, with the two flanking $R_z$ 's handling phase and the single $R_y$ handling the "tilt".

How do rotation gates make variational quantum algorithms trainable?

Because their angle dependence is a clean sinusoid, their gradients are exact and cheap. A rotation generator has just two eigenvalues, $\pm\tfrac12$ , so any circuit expectation $\langle H\rangle(\theta)$ is a pure sinusoid in each angle. That gives the parameter-shift rule: the exact derivative is $\partial_\theta\langle H\rangle=\tfrac12[\langle H\rangle_{\theta+\pi/2}-\langle H\rangle_{\theta-\pi/2}]$ , obtained by running the same circuit at two shifted angles — no finite-difference error, no ancillas. This turns the circuit into a differentiable function of its angles that classical optimisers can train, which is why variational ansätze are parametrised in rotation gates specifically rather than in arbitrary unitaries.