Module V·Article II·~1 min read

Self-Adjoint and Unitary Operators

Euclidean and Unitary Spaces

Turn this article into a podcast

Pick voices, format, length — AI generates the audio

Adjoint Operator

For a linear operator $A$ on a Euclidean space: the adjoint operator $A^*$ satisfies $(Av, w) = (v, A^*w)$ for all $v, w$.

In matrix form (orthonormal basis): $A^* = A^\mathsf{T}$ (the transposed matrix).

Self-Adjoint (Symmetric) Operators

$A = A^* \iff A = A^\mathsf{T}$ — a symmetric matrix.

Spectral theorem: A symmetric operator in a finite-dimensional Euclidean space has real eigenvalues and an orthonormal basis of eigenvectors (it can be diagonalized by an orthogonal transformation).

Proof: the eigenvalues are real (from $(Av, v) = \lambda(v, v) = (v, Av) = \bar{\lambda}(v, v)$); eigenvectors of different values are orthogonal.

Corollary: $A = QDQ^\mathsf{T}$ (spectral decomposition), where $Q$ is orthogonal, $D = \mathrm{diag}(\lambda_1, ..., \lambda_n)$.

$A$ is positive definite $\iff$ all $\lambda_i > 0 \iff A = BB^\mathsf{T}$ for some $B$ (Cholesky decomposition).

Unitary Spaces and Operators

A unitary space is a complex vector space with a Hermitian scalar product: $(u, v) = \bar{v}^\mathsf{T}u$, where $(v, u) = \overline{(u, v)}$.

Adjoint operator (Hermitian adjoint): $A^* = \bar{A}^\mathsf{T}$ (Hermitian conjugation, transposition + complex conjugation).

Hermitian operator: $A = A^*$ (a generalization of symmetric). Eigenvalues are real. In quantum mechanics observable quantities are Hermitian operators; their eigenvalues are measurement results.

Unitary operator: $UU^* = U^*U = I$ (generalization of orthogonal). Preserves norm and scalar product. Eigenvalues are on the unit circle.

Singular Value Decomposition (SVD)

Any matrix $A$ of size $m \times n$: $A = U \Sigma V^\mathsf{T}$, where $U$ ($m \times m$) and $V$ ($n \times n$) are orthogonal, $\Sigma$ is diagonal with $\sigma_1 \ge \sigma_2 \ge ... \ge 0$ (singular values).

SVD is a universal decomposition: rank = number of nonzero $\sigma$; the least squares solution via pseudoinverse; image compression (take the first $k$ singular values); PCA.

§ Act · what next