Hermitian Operators

Every electronic structure calculation run today — Hartree–Fock, density functional theory, coupled cluster, configuration interaction — diagonalizes a Hermitian (or in the generalized-eigenvalue-problem sense, Hermitian-with-respect-to-an-overlap-metric) Fock or Kohn–Sham matrix, and the entire numerical stability of quantum chemistry software rests on exploiting Hermiticity: Hermitian matrices have real eigenvalues and an orthogonal (or unitary) set of eigenvectors, which is why standard diagonalization routines (LAPACK's Hermitian eigensolvers, for instance) are faster and more numerically stable than routines for general matrices, and why molecular orbital coefficients come out as a genuinely orthonormal basis without extra enforcement. In spectroscopy, the reality of eigenvalues guaranteed by Hermiticity is the entire reason a UV-Vis, IR, or NMR peak position can be reported as a single real energy or frequency rather than a complex number — and the deviation from strict Hermiticity, engineered deliberately via complex absorbing potentials or non-Hermitian effective Hamiltonians, is now a standard computational tool for modeling autoionizing resonances, predissociation, and Auger decay, where the imaginary part of an eigenvalue directly encodes a physical lifetime through $\tau = \hbar/\Gamma$. Non-Hermitian quantum mechanics has also become an active research area for open quantum systems and PT-symmetric Hamiltonians, precisely because the standard theorem is understood well enough that chemists and physicists can identify exactly which idealizations are being relaxed and why. In magnetic resonance and molecular symmetry analysis, the fact that angular momentum operators are Hermitian underlies why $m_\ell$, $m_s$, and total angular momentum quantum numbers are always real half-integers or integers, directly setting the selection rules and Zeeman splitting patterns used in EPR and NMR spectrometers, and in materials science the same real-eigenvalue guarantee underlies band-structure calculations, where Hermitian Bloch Hamiltonians produce the real band energies plotted in every solid-state and photovoltaic materials paper.

Ask the tutor: Can you show me why the angular momentum operator $\hat{L}_z$ needs periodic boundary conditions specifically, and what would go wrong if we tried to treat $\phi$ like an ordinary Cartesian coordinate instead?

Start from the postulate that every observable $A$ in quantum mechanics is represented by a linear operator $\hat{A}$, and that a measurement of $A$ can only return one of the eigenvalues $a$ of the eigenvalue equation $\hat{A}\psi = a\psi$. This is already familiar from the study of Eigenvalue Equations and Quantization: quantization itself — discrete energy levels, discrete angular momentum — emerges from boundary conditions imposed on such eigenvalue equations. But nothing in the bare idea "operator has eigenvalues" forbids those eigenvalues from being complex numbers. A generic linear operator acting on a generic vector space happily produces complex eigenvalues (a rotation matrix in two dimensions has eigenvalues $\pm i$, for instance). Quantum mechanics needs an extra restriction to rule this out for physical operators, and Hermiticity is that restriction. The formal definition uses the inner product of functions. For two well-behaved (square-integrable, satisfying the boundary conditions of the problem) functions $f(x)$ and $g(x)$ on a domain, define

An operator $\hat{A}$ is Hermitian (some texts say self-adjoint, with a technical distinction about domains that does not matter at this level) if, for all admissible $f$ and $g$,

In words: you can let $\hat{A}$ act on the function to the right, or let it act on the function to the left (with the operator itself unchanged, only the bra taking the complex conjugate of the result), and the integral comes out the same either way. This is a nontrivial statement, because most differential operators, applied naively, do not have this symmetry — moving a derivative across an integral costs you a boundary term, via integration by parts, and Hermiticity is precisely the statement that these boundary terms vanish under the physical boundary conditions imposed on wavefunctions (decay at infinity, periodicity, or vanishing at a box edge). The reason this single algebraic condition is worth an entire article is that it does two jobs simultaneously. First, applied with $f=g=\psi$ an eigenfunction, it forces the eigenvalue $a$ to satisfy $a = a^{*}$, i.e., to be real — this is proven in full generality below and is the load-bearing result of the whole subject. Second, applied to two eigenfunctions with different eigenvalues, it forces those eigenfunctions to be orthogonal, which is the seed of the orthonormality machinery used everywhere in spectroscopy (transition selection rules, Fourier expansion of wavefunctions, perturbation theory sums). Both consequences follow from the same one-line definition, which is part of why the definition is so central to the subject rather than a technicality to memorize and forget.

The chemically important operators — position $\hat{x}$, momentum $\hat{p}=-i\hbar\, d/dx$, kinetic energy $\hat{T} = -\dfrac{\hbar^2}{2m}\dfrac{d^2}{dx^2}$, the Hamiltonian $\hat{H} = \hat{T}+\hat{V}(x)$ for real potential $V$, and the angular momentum operators $\hat{L}_x, \hat{L}_y, \hat{L}_z$ — are all Hermitian on the function spaces relevant to bound states. It is worth seeing concretely why momentum needs the factor of $i$ in its definition: without it, $d/dx$ alone is not Hermitian (integration by parts on a first derivative picks up a sign flip), but $-i\, d/dx$ exactly compensates that sign flip with the complex conjugation built into the inner product, restoring Hermiticity. This is not a cosmetic choice — it is forced by the requirement that momentum be a real, measurable quantity.

The central defining relation, restated for clarity and for later reference, is

where: $\hat{A}$ is the operator being tested, $f,g$ are any two functions satisfying the boundary conditions of the problem (normalizable, or periodic, or vanishing at box walls), and the star denotes complex conjugation. Reading: this equation says the operator can be transferred across the inner product from acting on the right function to acting on the left function without changing the value of the integral — this transferability is the entire content of "Hermitian." The payoff — real eigenvalues — is captured by

where: $\psi$ is a normalized eigenfunction of $\hat{A}$, $a$ is the corresponding eigenvalue, obtained by setting $f=g=\psi$ in the Hermiticity condition. Reading: real eigenvalues are not assumed — they are derived from Hermiticity, which is why physicists insist observables be represented by Hermitian operators in the first place. A second consequence, foreshadowing eigenfunction structure, is orthogonality for unequal eigenvalues:

where: $\psi_1,\psi_2$ are eigenfunctions of the same Hermitian operator with distinct eigenvalues $a_1 \ne a_2$. Reading: two states of a Hermitian observable with different measurable values are automatically orthogonal — this is the seed idea developed in full under Orthonormality of Eigenfunctions. Finally, the practical criterion used to verify Hermiticity for a differential operator on an interval $[a,b]$ is

where: the right-hand side is what remains after one or more integrations by parts. Reading: Hermiticity holds if and only if this boundary term vanishes — which is why the boundary conditions imposed on the wavefunction (decay at infinity, node at an infinite wall, periodicity on a ring) are not bookkeeping details but the actual mechanism enforcing real, measurable eigenvalues.

Example 1 — Foundational: checking Hermiticity for $\hat{p}=-i\hbar\,d/dx$ on a finite box, with unequal, non-round test functions. To make the boundary-term mechanism completely concrete, work on a finite interval $x\in[0,L]$ with $L = 3.70$ Å (an off-round, asymmetric choice on purpose), using the particle-in-a-box eigenfunctions

i.e., the $n=2$ and $n=5$ box states — deliberately unequal quantum numbers, not a symmetric $n=n$ case, so that the calculation actually tests the mechanism rather than trivially cancelling. Compute the left-hand side of the Hermiticity test directly:

Using the product-to-sum identity $\sin A\cos B = \tfrac12[\sin(A+B)+\sin(A-B)]$ with $A = 2\pi x/L$, $B=5\pi x/L$:

Each term integrates to $-\frac{L}{k\pi}\big[\cos(k\pi x/L)\big]_0^L = -\frac{L}{k\pi}\left[(-1)^k - 1\right]$. For $k=7$ (odd): $-\frac{L}{7\pi}(-1-1) = \frac{2L}{7\pi}$. For $k=3$ (odd): $\frac{2L}{3\pi}$. So the bracket gives

Therefore

and

Now compute the right side, $\int_0^L (\hat{p}f)^{*}g\,dx$. Since $\hat{p}f = -i\hbar\,df/dx$, its conjugate is $(\hat{p}f)^* = +i\hbar\,df/dx$ (as $f$ is real), so

By the same product-to-sum identity (now $\cos A\sin B = \tfrac12[\sin(A+B)-\sin(A-B)]$, $A=2\pi x/L,\ B=5\pi x/L$), the integral equals $\tfrac12\left(\frac{2L}{7\pi}+\frac{2L}{3\pi}\right) = \frac{L}{7\pi}+\frac{L}{3\pi} = \frac{10L}{21\pi}$. Then

so

Both sides equal $\dfrac{40\,i\hbar}{21L}$, confirming the Hermiticity condition for this specific, deliberately asymmetric pair of box states — and note that at no point did $f=0$ and $g=0$ at the endpoints get invoked inside the calculation; the box eigenfunctions automatically satisfy $\psi(0)=\psi(L)=0$, which is exactly the boundary condition that made this equality possible when the calculation is redone via integration by parts instead of direct evaluation (the two routes must and do agree).

Limiting case check. As $L\to\infty$ with fixed quantum numbers scaled so the wavevectors $k=n\pi/L$ stay fixed, both integrals scale as $1/L$ and vanish, consistent with free-particle plane waves not being individually normalizable — the finite-box result smoothly connects to the free-particle (delta-function-normalized) limit.

Example 2 — Applied: is the "naive" one-dimensional Hamiltonian with a linear ramp potential Hermitian, checked with a concrete asymmetric potential value? Consider an electron in a region with a uniform static electric field, giving potential energy $V(x) = eEx$ for $x\in[0,L]$ with field $E = 2.15\times10^{7}\ \text{V/m}$ and box width $L = 1.20$ nm (values chosen off-round, as in a Stark-effect-in-a-box problem). The Hamiltonian is

Confirm Hermiticity by checking each piece. The kinetic term is Hermitian by the general proof above, valid for any function vanishing at the box walls $x=0,L$ regardless of the potential present — the boundary term argument in the kinetic-energy derivation only used $f=g=0$ at the endpoints, never anything about $V(x)$. The potential term is multiplication by the real number $eEx$ at each point $x$:

using that $x$ and $eE$ are real numbers so $(eEx f)^* = eEx f^*$. This holds termwise for any real-valued $V(x)$ — linear, quadratic, Coulombic, Morse — without any integration by parts or boundary term at all, because multiplication operators commute past complex conjugation trivially when the multiplying function is real. So $\hat{H} = \hat{T}+\hat{V}$ is Hermitian, and its eigenvalues — the Stark-shifted energy levels of the electron in this field — are guaranteed real. As a numerical sanity check on the field strength itself: the potential energy accumulated across the box is

a modest perturbation compared to the unperturbed particle-in-a-box ground-state energy $E_1 = h^2/8m_eL^2 \approx 0.261$ eV for this box width — confirming this is a sensible, physically realizable Stark perturbation problem, and that the Hamiltonian written down is a legitimate real-valued, Hermitian operator whose real eigenvalues can genuinely be measured spectroscopically.

Example 3 — Exam-level trap: an operator that looks non-Hermitian until the boundary conditions are used correctly. A common exam trick is to present the operator $\hat{A} = i\dfrac{d}{dx}$ (note: no $\hbar$, and importantly no minus sign, as a deliberately altered form to test whether the student checks Hermiticity by rote or by calculation) on the domain of functions on a ring of circumference $\ell = 8.40$ Å with periodic boundary conditions $\psi(x+\ell) = \psi(x)$, and ask whether $\hat{A}$ is Hermitian, and if so, whether its eigenvalues are real. Take two periodic functions $f,g$ on $[0,\ell]$ with $f(0)=f(\ell)$, $g(0)=g(\ell)$. Compute

The boundary term is $i\left[f^{*}(\ell)g(\ell) - f^{*}(0)g(0)\right]$, which vanishes not because $f,g\to0$ (they do not, on a ring — this is the trap) but because periodicity gives $f^*(\ell)=f^*(0)$ and $g(\ell)=g(0)$, so the bracketed difference is exactly zero regardless of the specific values of $f(0)$ and $g(0)$. So

where the middle step used $\left(i\,df/dx\right)^{*} = -i\,df^{*}/dx$. The two sides match, so $\hat{A}=i\,d/dx$ is Hermitian on the periodic ring — despite lacking both the $\hbar$ and the customary minus sign that $\hat{p}=-i\hbar\,d/dx$ carries, because Hermiticity only cares about the structural pairing of the imaginary unit with the derivative and the vanishing of the boundary term, not about the numerical prefactor (any real, nonzero prefactor multiplying a Hermitian operator gives another Hermitian operator). Its eigenfunctions are $\psi_n(x) = \dfrac{1}{\sqrt{\ell}}e^{2\pi i n x/\ell}$ for integer $n$ (chosen to satisfy periodicity), with eigenvalues found from $\hat{A}\psi_n = i\cdot\dfrac{2\pi i n}{\ell}\psi_n = -\dfrac{2\pi n}{\ell}\psi_n$. Take $n=3$ on the ring of circumference $\ell = 8.40$ Å:

manifestly a real number, exactly as the general theorem demands. The trap in this problem is not the algebra — it is realizing that "vanishing boundary term" does not always mean "function vanishes at the boundary"; on a ring it means "function returns to its starting value," and periodicity supplies that just as effectively as decay-at-infinity does for a free particle.

The mathematical structure underlying Hermitian operators is named for Charles Hermite, a French mathematician who never worked on quantum mechanics himself — he died in 1901, a quarter-century before Schrödinger's equation existed — but whose work on quadratic forms, orthogonal polynomials, and self-adjoint transformations in the 1850s–1870s supplied exactly the linear-algebraic machinery that Dirac, von Neumann, and others needed when they formalized quantum theory in the late 1920s. Hermite's own preoccupations were classical: he proved the transcendence of $e$ in 1873, studied elliptic functions, and introduced the polynomials that now bear his name in the context of probability theory and interpolation, long before anyone suspected they would become the exact eigenfunctions of the quantum harmonic oscillator. The physical use of Hermiticity — insisting that observables be represented by self-adjoint operators specifically so that measured quantities come out real — is due to the architects of matrix and wave mechanics: Werner Heisenberg's matrix formulation implicitly required Hermitian matrices for observable quantities, and Paul Dirac's 1930 textbook The Principles of Quantum Mechanics made the operator formalism, including the explicit real-eigenvalue theorem, the language of the field. John von Neumann's 1932 mathematical treatise then supplied the fully rigorous Hilbert-space treatment (including the subtle distinction between Hermitian and truly self-adjoint operators on infinite-dimensional spaces, a distinction that rarely bites in standard chemistry problems but matters for some singular potentials).