Hermitian Operators: Real Spectra, Orthogonal Modes
~6 min read · Tier toggle changes depth (T1 plain · T2 standard · T3 rigour)
You should be comfortable with the definition of the inner product in a complex vector space, the concept of the adjoint (Hermitian conjugate) of an operator, and basic eigenvalue problems. Familiarity with the Dirac bra-ket notation is assumed.
Why this matters
Hermitian operators are the mathematical guarantee that physical measurements yield real numbers and that distinct quantum states can be cleanly separated.
In classical mechanics, normal modes of vibration are orthogonal; in quantum mechanics, observables are represented by operators whose eigenvalues are the possible measurement outcomes. If these operators were not Hermitian, we would face the absurdity of measuring a complex energy or finding that two distinct stationary states are not orthogonal, making probability calculations incoherent. This lecture establishes the rigorous link between the algebraic property of self-adjointness and these physical necessities.
The Intuition: Symmetry and Reality
Imagine a system that looks the same when you swap the roles of "input" and "output" in a specific mathematical sense. In linear algebra, this symmetry is captured by the Hermitian condition. If an operator is Hermitian, it behaves like a pure stretching along mutually perpendicular axes, with real stretch factors and no rotation or hidden complex phase that would twist the geometry of the space. This symmetry ensures that when the operator acts on a state, the resulting "length" or "magnitude" changes in a way that is strictly real. Furthermore, because the operator is symmetric, it cannot mix two distinct "directions" (eigenvectors) in a way that creates a non-zero overlap. Distinct modes of vibration, or distinct quantum states, remain perfectly orthogonal, allowing us to decompose any arbitrary state into a clean sum of these independent components without cross-talk.
Real Eigenvalues
The first and most critical consequence of the Hermitian property is that all eigenvalues are real numbers. Let \(\hat{A}\) be a Hermitian operator, such that \(\hat{A}^\dagger = \hat{A}\). Let \(|\psi\rangle\) be an eigenvector of \(\hat{A}\) with eigenvalue \(\lambda\). By definition:
We take the inner product of both sides with \(|\psi\rangle\):
Since \(\hat{A}\) is Hermitian, \(\langle\psi|\hat{A}|\psi\rangle\) is equal to its own complex conjugate. Let \(z = \langle\psi|\hat{A}|\psi\rangle\). Then \(z = z^*\). A complex number that is equal to its own conjugate must be real. Therefore, \(\lambda \langle\psi|\psi\rangle\) is real. Since \(\langle\psi|\psi\rangle\) is a positive real number (the norm squared), \(\lambda\) itself must be real. This result is fundamental: it ensures that any observable quantity represented by a Hermitian operator yields a real-valued measurement result.
Orthogonality of Distinct Eigenvectors
The second major consequence is that eigenvectors corresponding to distinct eigenvalues are orthogonal. Let \(|u\rangle\) and \(|v\rangle\) be eigenvectors of \(\hat{A}\) with distinct eigenvalues \(\lambda_u\) and \(\lambda_v\), where \(\lambda_u \neq \lambda_v\).
Consider the inner product \(\langle u|\hat{A}|v\rangle\). We can evaluate this in two ways. First, acting on \(|v\rangle\):
Second, using the Hermitian property \(\hat{A}^\dagger = \hat{A}\), we move the operator to the bra side:
Since we already proved that eigenvalues of Hermitian operators are real, \(\lambda_u^* = \lambda_u\). Equating the two expressions:
Because \(\lambda_u \neq \lambda_v\), the term \((\lambda_v - \lambda_u)\) is non-zero. Therefore, we must have \(\langle u|v\rangle = 0\). This orthogonality is crucial for the completeness of the eigenbasis, allowing any state to be expanded uniquely in terms of these orthogonal modes.
Physical Incarnations
In quantum mechanics, the Hamiltonian operator \(\hat{H}\) is Hermitian. This guarantees that energy eigenvalues are real, which is a prerequisite for the time-evolution operator \(e^{-i\hat{H}t/\hbar}\) to be unitary (preserving probability). Similarly, the momentum operator \(\hat{p} = -i\hbar \frac{d}{dx}\) is Hermitian (with appropriate boundary conditions), ensuring that momentum measurements are real. The orthogonality of eigenstates ensures that if a system is prepared in an energy eigenstate, it will not spontaneously transition to a different energy eigenstate due to "overlap" in the Hilbert space; the states are fundamentally distinct and separable.
The derivation
Properties of Hermitian Operators
If † = Â, then every eigenvalue of  is real, and eigenvectors belonging to distinct eigenvalues are orthogonal.
A common pitfall is assuming that any operator with real eigenvalues is Hermitian. This is false. An operator can have real eigenvalues but still be non-Hermitian (e.g., a non-symmetric matrix with real eigenvalues). The Hermitian condition is stronger: it requires the operator to be equal to its own adjoint. This distinction is vital when dealing with non-orthogonal bases or defective matrices (matrices that do not have a complete set of eigenvectors). Only Hermitian operators guarantee both real spectra and a complete orthogonal eigenbasis. C
Worked example
Problem. Consider the \(2 \times 2\) matrix \(M = \begin{pmatrix} 1 & i \\ -i & 1 \end{pmatrix}\). Determine if \(M\) is Hermitian. If it is, find its eigenvalues and eigenvectors, and verify that the eigenvectors are orthogonal.
- Check Hermiticity: Compute \(M^\dagger\). The conjugate transpose is obtained by taking the complex conjugate and then the transpose. \[ M^\dagger = \left( \begin{pmatrix} 1 & i \\ -i & 1 \end{pmatrix}^* \right)^T = \begin{pmatrix} 1 & -i \\ i & 1 \end{pmatrix}^T = \begin{pmatrix} 1 & i \\ -i & 1 \end{pmatrix} = M \] Since \(M^\dagger = M\), the matrix is Hermitian.
- Find Eigenvalues: Solve \(\det(M - \lambda I) = 0\). \[ \det \begin{pmatrix} 1-\lambda & i \\ -i & 1-\lambda \end{pmatrix} = (1-\lambda)^2 - (i)(-i) = (1-\lambda)^2 - 1 = 0 \] \[ (1-\lambda)^2 = 1 \implies 1-\lambda = \pm 1 \] Case 1: \(1-\lambda = 1 \implies \lambda_1 = 0\). Case 2: \(1-\lambda = -1 \implies \lambda_2 = 2\). Both eigenvalues are real, consistent with the Hermitian property.
- Find Eigenvectors: For \(\lambda_1 = 0\): \[ \begin{pmatrix} 1 & i \\ -i & 1 \end{pmatrix} \begin{pmatrix} x \\ y \end{pmatrix} = \begin{pmatrix} 0 \\ 0 \end{pmatrix} \implies x + iy = 0 \implies x = -iy \] Let \(y = 1\), then \(x = -i\). Eigenvector \(|v_1\rangle = \begin{pmatrix} -i \\ 1 \end{pmatrix}\). For \(\lambda_2 = 2\): \[ \begin{pmatrix} -1 & i \\ -i & -1 \end{pmatrix} \begin{pmatrix} x \\ y \end{pmatrix} = \begin{pmatrix} 0 \\ 0 \end{pmatrix} \implies -x + iy = 0 \implies x = iy \] Let \(y = 1\), then \(x = i\). Eigenvector \(|v_2\rangle = \begin{pmatrix} i \\ 1 \end{pmatrix}\).
- Verify Orthogonality: Compute \(\langle v_1 | v_2 \rangle\). \[ \langle v_1 | = \begin{pmatrix} i & 1 \end{pmatrix} \] \[ \langle v_1 | v_2 \rangle = \begin{pmatrix} i & 1 \end{pmatrix} \begin{pmatrix} i \\ 1 \end{pmatrix} = i(i) + 1(1) = -1 + 1 = 0 \] The eigenvectors are orthogonal.
Units check. The matrix elements are dimensionless in this abstract example, so the eigenvalues are dimensionless. The orthogonality condition is a scalar result equal to zero, which is dimensionally consistent.
Three quick tests
- Is the matrix \( \begin{pmatrix} 1 & 2 \\ 2 & 1 \end{pmatrix} \) Hermitian?
Answer
Yes. It is real and symmetric, so \( A^\dagger = A^T = A \). - Can a Hermitian operator have a complex eigenvalue?
Answer
No. All eigenvalues of a Hermitian operator are real. - If two eigenvectors of a Hermitian operator correspond to the same eigenvalue, are they necessarily orthogonal?
Answer
No. They are not necessarily orthogonal, but they can be chosen to be orthogonal via the Gram-Schmidt process. The theorem only guarantees orthogonality for distinct eigenvalues.
Problem set for PU-104 →
Each derivation in this unit ends with five graded problems with full solutions.
Lecture notes drafted with a local model and checked line by line by the editor; every number in the worked example was recomputed.