Hermitian Matrices

A physicist wants a matrix to stand in for a measurable quantity — an energy, a position, a spin. Whatever the machinery inside, one thing is non-negotiable: when you read the dial, the number that comes out is real. Which matrices are guaranteed to have only real "readings" (eigenvalues), even when their entries are complex? The answer is a beautiful and sharply-defined family: the Hermitian matrices. They are the complex world's version of a symmetric matrix — and they turn out to be exactly the right tool for the job.

For a real matrix, symmetry means A^T = A: flip it across the diagonal and nothing changes. Over the complex numbers, plain transposing is the wrong move, because the natural partner of the transpose also conjugates every entry. That combined operation — transpose and conjugate — is the conjugate transpose (or adjoint), written A^\dagger and read "A-dagger":

A^\dagger = \overline{A^{\,T}} = \overline{A}^{\,T}, \qquad (A^\dagger)_{ij} = \overline{A_{ji}}.

A matrix that is left unchanged by this operation — equal to its own conjugate transpose — is Hermitian:

A^\dagger = A.

Because conjugating a real number does nothing, a real Hermitian matrix is just an ordinary symmetric matrix. Hermitian is the honest complex generalisation of "symmetric".

The shape of a Hermitian matrix

The rule A_{ij} = \overline{A_{ji}} pins down exactly what one looks like. On the diagonal, an entry is its own mirror image, so A_{ii} = \overline{A_{ii}} — and the only numbers equal to their own conjugate are the real ones. Off the diagonal, each entry must be the conjugate of its partner across the diagonal. So a Hermitian matrix has a real diagonal, with conjugate pairs reflected across it:

A = \begin{bmatrix} 2 & 3 - i & -4i \\ 3 + i & 5 & 1 \\ 4i & 1 & 0 \end{bmatrix}.

Check it: the diagonal 2, 5, 0 is real; the pair 3-i / 3+i are conjugates, and so are -4i / 4i. Every quantum observable is built from matrices exactly like this — the three Pauli matrices are the smallest famous examples.

Worked example: is it Hermitian?

Take the Pauli-Y matrix and test it. Transpose first, then conjugate every entry:

Y = \begin{bmatrix} 0 & -i \\ i & 0 \end{bmatrix}, \quad Y^T = \begin{bmatrix} 0 & i \\ -i & 0 \end{bmatrix}, \quad Y^\dagger = \overline{Y^T} = \begin{bmatrix} 0 & -i \\ i & 0 \end{bmatrix} = Y.

The conjugate transpose comes back to Y itself, so Y is Hermitian — even though it is stuffed with imaginary numbers and its diagonal happens to be zero (zero is real, so that is fine).

Worked example: the eigenvalues really are real

Take the little Hermitian matrix A = \begin{bmatrix} 2 & 3-i \\ 3+i & 5 \end{bmatrix} and find its eigenvalues. The off-diagonal product is (3-i)(3+i) = 9 - i^2 = 10, a real number, so the characteristic equation is

\det(A - \lambda I) = (2-\lambda)(5-\lambda) - 10 = \lambda^2 - 7\lambda = \lambda(\lambda - 7) = 0.

The eigenvalues are \lambda = 0 and \lambda = 7 — both real, exactly as the theorem promises, with no imaginary part in sight. This is not luck: it happens for every Hermitian matrix.

the conjugate transpose is A^\dagger = \overline{A}^{\,T} — transpose and conjugate;
A is Hermitian when A^\dagger = A;
the diagonal entries are real, and mirror entries are conjugates (A_{ij} = \overline{A_{ji}});
every eigenvalue is real, and eigenvectors for distinct eigenvalues are orthogonal — the complex spectral theorem.

In quantum mechanics, every measurable quantity — energy, momentum, a component of spin — is represented by a matrix (an "operator"), and the possible results of measuring it are that matrix's eigenvalues. Measured numbers are real: an energy meter never reads 3 + 2i joules. So the operator must be one whose eigenvalues are guaranteed real no matter what — and that is precisely a Hermitian matrix. The spectral theorem throws in a bonus: its eigenvectors are orthogonal, which become the distinct, perfectly distinguishable outcomes of the measurement. Hermitian isn't a convenient choice here; it's forced by physics.

Over the complex numbers, "flip across the diagonal" (A^T = A) is not enough — you must conjugate too. The matrix M = \begin{bmatrix} 0 & i \\ i & 0 \end{bmatrix} is symmetric (M^T = M), yet M^\dagger = \begin{bmatrix} 0 & -i \\ -i & 0 \end{bmatrix} \ne M, so it is not Hermitian. Tell-tale sign: its diagonal here is real by accident, but a symmetric matrix is allowed a complex diagonal, and a Hermitian one never is. Always take the conjugate, not just the transpose.