Unitary and Orthogonal Operators

Unitarity is a daily working constraint across physics and computation. In quantum computing every gate is a unitary matrix — the Hadamard and $SU(2)$ gates of Examples 2 and 3 are literal hardware operations, and an algorithm is a product of unitaries steering a state around the unit sphere without losing a drop of probability; unitary synthesis asks how to decompose an arbitrary $U \in SU(2^n)$ into available gates. In metrology, $SO(3)$ governs the attitude estimation that keeps satellites and drones pointed, best done through unit quaternions, the double cover $SU(2) \to SO(3)$. In large-scale computation, orthogonal transforms are prized because their condition number is exactly one: Householder reflections and Givens rotations power the QR eigenvalue algorithm, and the discrete Fourier transform — unitary up to normalization — underlies the FFT. And in fundamental physics the unitarity of the S-matrix — that scattering conserve total probability — is what forced the prediction, and discovery, of the Higgs boson from the electroweak amplitudes.

Picture a rigid tripod of three mutually perpendicular unit arrows glued at the origin. An orthogonal operator on $\mathbb{R}^3$ is any way of re-orienting that tripod — spinning it, flipping it — without bending an arrow or changing an angle between two of them. That indestructible rigidity is precisely $Q^\mathsf{T} Q = I$: the columns of $Q$ are the images of the tripod's arrows, and the equation says those images are still orthonormal. A unitary operator does the same for complex "arrows", with one extra freedom — each complex coordinate may also pick up a phase $e^{i\theta}$, an internal rotation invisible to real geometry but essential to quantum amplitudes. The first figure makes the two-dimensional version concrete with real numbers.

The condition $\langle Ux, Uy\rangle = \langle x, y\rangle$ says the operator moves the whole space the way a rotation does — no direction favored, no pair sheared, no volume inflated. This is why unitarity is the natural home of conservation: preserved lengths mean the squared norm of a quantum state is frozen, so total probability cannot leak away. The deepest fact is spectral: because $U$ commutes with $U^\dagger$ it is diagonalizable in an orthonormal eigenbasis, and in each eigendirection it merely multiplies by a unit-modulus $e^{i\theta}$. The whole operator is a bundle of independent rotations, one per eigenplane, every eigenvalue pinned to the unit circle. The second figure shows that circle.

The two definitions look almost identical, and that is the point: an orthogonal operator is the real specialization of a unitary one, the adjoint $\dagger$ (conjugate-transpose) collapsing to the transpose $\mathsf{T}$ for real entries. We state both, then unpack the geometric fact they encode.

where: $U$ — operator on a complex inner-product space $V$; $U^\dagger$ — Hermitian adjoint, $\langle U^\dagger x, y\rangle = \langle x, U y\rangle$; $I$ — identity. Reading: the adjoint is the inverse — undoing $U$ costs only a conjugate-transpose.

where: $Q$ — operator on a real inner-product space; $Q^\mathsf{T}$ — transpose, the real adjoint. Reading: identical to (1); everything below specializes by $\dagger \to \mathsf{T}$. The whole geometric meaning is compressed into one line: $U$ preserves the inner product, hence length and angle. This is the equation to remember — the algebra $U^\dagger U = I$ and the geometry below are logically equivalent.

where: $\langle\cdot,\cdot\rangle$ — inner product on $V$; $\lVert x\rVert = \sqrt{\langle x, x\rangle}$ — induced norm. Reading: norm preservation is the case $y = x$; via polarization the two are equivalent over $\mathbb{C}$. In components (orthonormal basis), $U^\dagger U = I$ becomes a statement about the columns of $U$: they form an orthonormal set — the most practical hand-check for unitarity.

where: $(U)_{ik}$ — matrix entry in an orthonormal basis; $\overline{(\cdot)}$ — complex conjugate; $\delta_{k\ell}$ — Kronecker delta. Reading: column $k$ dotted with column $\ell$ gives $\delta_{k\ell}$ — columns and rows are orthonormal frames. Norm preservation constrains the spectrum: an eigenvector's length cannot change, so its eigenvalue cannot have modulus other than one.

where: $\lambda$ — eigenvalue; $v \neq 0$ — eigenvector; $\theta$ — real phase angle. Reading: a real orthogonal operator therefore admits only $\pm 1$ as real eigenvalues, the rest in conjugate pairs $e^{\pm i\theta}$. The determinant inherits the same discipline, and separates the two connected pieces of the real orthogonal group.

where: $\det$ — determinant. Reading: $\det Q = +1$ is a proper rotation in $SO(n)$; $\det Q = -1$ is an improper motion (rotation plus reflection), and the two never connect continuously. Because $U^\dagger U = U U^\dagger$, $U$ is normal, and the spectral theorem gives an orthonormal eigenbasis: $U$ is a sum of phases times orthogonal projectors.

where: $P_k$ — orthogonal projector onto the $k$-th eigenspace; $e^{i\theta_k}$ — unit-modulus eigenvalue. Reading: $U$ is a bundle of independent phase rotations. Two operational forms close the set. First, a unitary is the change between orthonormal frames, a geometry-preserving similarity.

where: $A$ — operator in the original orthonormal basis; $A'$ — the same operator in the new frame; $U$ — unitary whose columns are the new basis vectors. Reading: diagonalizing a Hermitian observable means choosing $U$ so $A'$ is diagonal. Second, the canonical building block of every real orthogonal matrix is the $2\times 2$ rotation block, alongside $\pm 1$ entries.

where: $\theta$ — rotation angle; $R(\theta) \in SO(2)$. Reading: stacked with $\pm 1$s, this block reconstructs any rotation or reflection in any dimension.

Pushing the definition to its edges shows what unitarity guarantees — and, in the last row, where finite-dimensional intuition breaks.

1 · A rotation is orthogonal — check every claim with numbers (fully worked)

Take the plane rotation by $\theta = 30°$, $R = \begin{pmatrix} \cos 30° & -\sin 30° \\ \sin 30° & \cos 30° \end{pmatrix}$. Verify orthogonality, the determinant, and the eigenvalues.

Step 1. Insert $\cos 30° = 0.8660$, $\sin 30° = 0.5000$, so $R = \begin{pmatrix} 0.8660 & -0.5000 \\ 0.5000 & 0.8660 \end{pmatrix}$ with columns $c_1 = (0.8660, 0.5000)$, $c_2 = (-0.5000, 0.8660)$.

Step 2. Check $R^\mathsf{T} R = I$ column by column: $c_1 \cdot c_1 = 0.7500 + 0.2500 = \mathbf{1.0000}$, identically $c_2 \cdot c_2 = 1.0000$, and the cross term $c_1 \cdot c_2 = -0.4330 + 0.4330 = \mathbf{0}$. Columns orthonormal, so $R^\mathsf{T} R = I$ — equation (4) made arithmetic.

Step 3. Determinant: $\det R = 0.7500 + 0.2500 = \mathbf{+1}$. Positive determinant means a proper rotation in $SO(2)$ — no reflection.

Step 4. Eigenvalues from $\lambda^2 - (\operatorname{tr} R)\lambda + \det R = 0$, i.e. $\lambda^2 - 2(0.8660)\lambda + 1 = 0$. The discriminant is $3 - 4 = -1 < 0$, giving $\lambda = 0.8660 \pm 0.5000\, i = \mathbf{e^{\pm i 30°}}$, both on the unit circle ($|\lambda| = 1$) as equation (5) demands.

Step 5. A real rotation has no real eigenvector — nothing is left pointing the same way — which is why the eigenvalues are the complex pair $e^{\pm i\theta}$ acting on the single rotation plane. $R^\mathsf{T} R = I$, $\det R = +1$, eigenvalues $e^{\pm i30°} = 0.866 \pm 0.500\, i$ — a proper rotation of $SO(2)$.

2 · The Hadamard gate — orthogonal and unitary, but a reflection

The Hadamard gate is $H = \tfrac{1}{\sqrt 2}\begin{pmatrix} 1 & 1 \\ 1 & -1 \end{pmatrix}$. Show it is both orthogonal and unitary, find its determinant and eigenvalues, and identify the motion.

Step 1. $H$ is real symmetric, so $H^\mathsf{T} = H^\dagger = H$. Then $H^\mathsf{T} H = H^2 = \tfrac12\begin{pmatrix} 2 & 0 \\ 0 & 2 \end{pmatrix} = I$, so $H$ is orthogonal, and being real with $H^\dagger H = I$ it is unitary too. It is also an involution: $H^2 = I$, so $H = H^{-1}$.

Step 2. Determinant: $\det H = \tfrac12\big[(1)(-1) - (1)(1)\big] = \mathbf{-1}$. Negative determinant means $H$ is improper — not a pure rotation.

Step 3. Eigenvalues: from $H^2 = I$ every $\lambda^2 = 1$, so $\lambda = \pm 1$; the trace $\operatorname{tr} H = 0$ fixes them as $\{+1, -1\}$, consistent with $\det H = -1$. Both have $|\lambda| = 1$, as required.

Step 4. Identify the motion. An improper $2\times 2$ orthogonal matrix with eigenvalues $+1, -1$ is a reflection (the $+1$ eigenvector is the mirror axis, the $-1$ is flipped). Matching $H$ to the reflection form $\begin{pmatrix} \cos 2\varphi & \sin 2\varphi \\ \sin 2\varphi & -\cos 2\varphi \end{pmatrix}$ gives $\cos 2\varphi = 1/\sqrt 2$, so $\varphi = 22.5°$. $H$ is orthogonal and unitary with $\det = -1$ and eigenvalues $\pm 1$ — a reflection about the $22.5°$ axis, which is why it maps $|0\rangle, |1\rangle$ to the two equal superpositions.

3 · A genuinely complex unitary — a qubit rotation in $SU(2)$

Consider $U = \tfrac{1}{\sqrt 2}\begin{pmatrix} 1 & i \\ i & 1 \end{pmatrix}$. Confirm it is unitary, find its determinant and eigenvalues, and connect it to a Hermitian generator.

Step 1. Adjoint (conjugate-transpose): $U^\dagger = \tfrac{1}{\sqrt 2}\begin{pmatrix} 1 & -i \\ -i & 1 \end{pmatrix}$. Then $U^\dagger U = \tfrac12\begin{pmatrix} 1+1 & i - i \\ -i + i & 1 + 1 \end{pmatrix} = I$, so $U$ is unitary — the conjugation is essential; the plain transpose would not give $I$.

Step 2. Determinant: $\det U = \tfrac12\big[(1)(1) - (i)(i)\big] = \mathbf{+1}$. Unitary with $\det = 1$ places $U$ in $SU(2)$.

Step 3. Eigenvalues: trace $\sqrt 2$, $\det = 1$, so $\lambda^2 - \sqrt 2\,\lambda + 1 = 0$ gives $\lambda = \tfrac{1}{\sqrt2}(1 \pm i) = \mathbf{e^{\pm i\pi/4}}$, on the unit circle.

Step 4. Connect to a generator. With $X = \begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}$ (Hermitian), $e^{i\alpha X} = \cos\alpha\, I + i\sin\alpha\, X$; at $\alpha = \pi/4$ this is $\tfrac{1}{\sqrt2}(I + iX) = U$. So $U = e^{i(\pi/4)X}$ — a Bloch-sphere rotation generated by $X$, and $\det U = e^{i(\pi/4)\operatorname{tr} X} = 1$ confirms $SU(2)$. $U$ is unitary with $\det = 1$ and eigenvalues $e^{\pm i\pi/4}$; it equals $e^{i(\pi/4)X}$ — the recipe "unitary = exponential of $i$ times Hermitian".

Unitary and orthogonal operators are how pure linear algebra became the grammar of physical symmetry — a thread from Cauchy's spectral theorem, through Hermite's forms, to the moment quantum mechanics made unitarity synonymous with conservation of probability. No one drew that synthesis more sharply than Hermann Weyl.

Texts. Arfken and Weber, Mathematical Methods for Physicists, 7th ed., ch. 3 (matrices and orthogonal/unitary transformations) · Hassani, Mathematical Physics, 2nd ed., chs. on operators and the spectral decomposition · Stone and Goldbart, Mathematics for Physics, ch. on linear operators and Hilbert space · Nakahara, Geometry, Topology and Physics, 2nd ed., §4 on Lie groups $U(n)$, $SU(n)$, $O(n)$, $SO(n)$ · Byron and Fuller, Mathematics of Classical and Quantum Physics, chs. 3–5 · Halmos, Finite-Dimensional Vector Spaces, §§ on unitary and orthogonal transformations. Key results to keep in mind: the complex polarization identity rebuilds $\langle x, y\rangle$ from four norms; $|\lambda| = 1$ follows from $\lVert Uv\rVert = \lVert v\rVert$; $|\det U|^2 = \overline{\det U}\,\det U = 1$; $R(30°)$ has orthonormal columns, $\det = +1$, and eigenvalues $e^{\pm i30°} = 0.866 \pm 0.500\,i$; the Hadamard gate has $\det = -1$ and eigenvalues $\pm 1$; and $U = \tfrac{1}{\sqrt2}\big(\begin{smallmatrix} 1 & i \\ i & 1 \end{smallmatrix}\big)$ is in $SU(2)$ with eigenvalues $e^{\pm i\pi/4}$ and equals $e^{i(\pi/4)X}$.

Prerequisites. Inner-Product Spaces and Norms · Adjoint and Hermitian Operators · Orthogonality and the Gram–Schmidt Process · Linear Operators and Matrix Representations. Leads to. Diagonalization and the Spectral Decomposition · Change of Basis and Similarity Transformations · Eigenvalues, Eigenvectors, and the Characteristic Equation · Determinants and Their Properties.

Ask the tutor: Want to see why demanding a unitary S-matrix predicted the Higgs, or how unit quaternions in $SU(2)$ beat rotation matrices for spacecraft attitude — just ask.