Orthonormality of Eigenfunctions

Every electronic structure code — Gaussian, ORCA, Psi4, VASP, Q-Chem — relies on orthonormality at its computational core. Molecular orbitals produced by a Hartree–Fock or DFT calculation are, by construction, an orthonormal set (the self-consistent-field procedure explicitly enforces $\int \psi_i^*\psi_j\,d\tau = \delta_{ij}$ via Löwdin or Gram–Schmidt orthogonalization steps), which is precisely why a molecular orbital diagram can list orbital energies and occupations without worrying about cross-terms contaminating the total electronic energy. In spectroscopy, the orthonormality of vibrational and rotational eigenfunctions is what makes selection rules calculable: the transition dipole moment integral $\int \psi_f^* \hat{\mu} \psi_i \, d\tau$ is evaluated against an orthonormal vibrational basis (harmonic oscillator eigenfunctions), and it is exactly the near-orthogonality of these functions that produces the sharp propensity rules (like $\Delta v = \pm 1$ for the harmonic oscillator) used to interpret infrared and Raman spectra of real molecules. In NMR and EPR spectroscopy, spin eigenfunctions ($\alpha, \beta$ states, or the full basis of a coupled spin system) are orthonormal by the same theorem applied to the spin operators $\hat{S}_z$ and $\hat{S}^2$; this orthonormality underlies the entire product-basis and density-matrix formalism used to simulate multi-spin pulse sequences. In materials science, Bloch's theorem for electrons in a periodic crystal produces orthonormal Bloch eigenfunctions labeled by crystal momentum $k$ and band index $n$; band-structure diagrams and density-of-states calculations for semiconductors and battery materials are built entirely on this orthonormal eigenbasis. In drug design, orthonormal molecular orbital expansions underlie every quantum-mechanical docking and reactivity calculation (Fukui functions, frontier-orbital HOMO/LUMO reasoning) used to predict which site on a candidate molecule will react first. Even quantum computing error-correction codes and quantum information protocols lean on orthonormal eigenbases of Pauli operators to define measurement outcomes unambiguously — the same theorem, in a very different technological context.

Ask the tutor: How does orthonormality of eigenfunctions justify treating $|c_n|^2$ as a measurement probability in a superposition state?

Think of an operator's eigenfunctions as a set of "pure notes" that an instrument can produce. When you pluck a guitar string, the sound you hear is a superposition of harmonics — the fundamental, the first overtone, the second overtone, and so on. Each harmonic is a stationary vibrational pattern with a definite frequency, and critically, the harmonics do not "leak" into each other: the energy carried by the fundamental and the energy carried by the second overtone are cleanly separable and additive, never cross-contaminating. This separability is a direct physical manifestation of orthogonality. In quantum mechanics, the "harmonics" are eigenfunctions of a Hermitian operator (energy, angular momentum, position, whatever observable you are measuring), and the fact that they don't leak into each other is precisely the statement $\int \psi_m^*\psi_n\,d\tau = 0$ for $m \neq n$. Geometrically, orthonormal eigenfunctions behave exactly like an orthonormal basis of vectors in ordinary 3D space, such as $\hat{i}, \hat{j}, \hat{k}$. The overlap integral $\int \psi_m^*\psi_n\,d\tau$ is the direct analogue of the dot product $\hat{e}_m \cdot \hat{e}_n$. Just as any vector $\vec{v}$ can be resolved into components along $\hat{i}, \hat{j}, \hat{k}$ by dot products, $v_x = \vec{v}\cdot\hat{i}$, any well-behaved wavefunction $\Psi$ can be resolved into components along the eigenfunctions $\psi_n$ by the analogous integral, $c_n = \int \psi_n^*\Psi\,d\tau$. This is why physicists speak of "Hilbert space" as an infinite-dimensional vector space of functions: the language of vectors, dot products, and orthonormal bases transfers over almost without modification, with integration over all space playing the role of the dot product. The first figure below shows two particle-in-a-box eigenfunctions, $\psi_1$ and $\psi_2$, plotted on top of each other, with their product $\psi_1\psi_2$ shaded to show that the positive and negative lobes exactly cancel upon integration. The second figure shows the abstract picture: eigenfunctions as mutually perpendicular axes in function space, with an arbitrary state $\Psi$ resolved into its projections. The third figure sketches what happens with degeneracy — two eigenfunctions sharing one eigenvalue are not forced to be perpendicular, but Gram–Schmidt rotates them into an orthogonal pair without changing the physics.

The central object is the eigenvalue equation for a Hermitian operator $\hat{A}$ representing an observable:

where: $\hat{A}$ is a linear Hermitian operator, $\psi_n$ is its $n$-th eigenfunction, and $a_n$ is the corresponding real eigenvalue (guaranteed real by Hermiticity). Reading: this says that applying the operator to $\psi_n$ simply returns the same function scaled by the number $a_n$ — no distortion of shape, only a rescaling. Different $n$ generically give different, physically distinguishable measurement outcomes $a_n$. The orthogonality statement, proven below, is:

where: the integral runs over all space (all relevant coordinates), $\psi_m^*$ is the complex conjugate of $\psi_m$, and $d\tau$ is the appropriate volume element (e.g. $dx$ in 1D, $r^2\sin\theta\,dr\,d\theta\,d\phi$ in spherical coordinates). Reading: eigenfunctions tied to different measurable outcomes are geometrically perpendicular in function space; there is zero "overlap" between them. Combining orthogonality with the normalization condition $\int |\psi_n|^2 d\tau = 1$ (from Normalization of the Wavefunction) gives the compact orthonormality relation:

where: $\delta_{mn}$ is the Kronecker delta, equal to $1$ if $m = n$ and $0$ if $m \neq n$. Reading: this single equation packages two separate physical facts — every eigenfunction has unit "length" (normalization) and every pair of distinct eigenfunctions is perpendicular (orthogonality) — into one universally used symbol. Finally, if the set $\{\psi_n\}$ is complete (spans the full space of physically acceptable functions for the problem), any admissible wavefunction $\Psi$ can be expanded as:

where: $c_n$ are complex expansion coefficients, obtained by projecting $\Psi$ onto each basis function, exactly analogous to resolving a vector into orthogonal components. Reading: this is the master tool of quantum chemistry — it says any wavefunction, no matter how complicated, is just a weighted recipe of the operator's own eigenfunctions, and the recipe's weights are computable by a simple integral rather than by trial and error.

Example 1 — Foundational: overlap integral for a particle in a box, verified with real numbers. The normalized eigenfunctions for a particle of mass $m$ in an infinite square well of length $L$ are

with distinct eigenvalues $E_n = n^2\hbar^2\pi^2/(2mL^2)$ for every distinct positive integer $n$. Take an asymmetric, non-round case: $L = 3.70\ \text{Å}$, and check $\psi_2$ against $\psi_5$ (an arbitrary unequal pair, chosen precisely because it is not the "obvious" symmetric $n=1,2$ pair used in the figure). The overlap integral is

Use the product-to-sum identity $\sin A \sin B = \tfrac{1}{2}[\cos(A-B) - \cos(A+B)]$ with $A = 2\pi x/L$, $B = 5\pi x/L$:

Each term integrates to a sine function evaluated at the endpoints:

for any nonzero integer $k$, since $\sin(k\pi) = 0$ exactly. Both $k = -3$ and $k = 7$ are nonzero integers, so both terms vanish identically:

The numeric value of $L = 3.70$ Å never even enters the final answer — a structural confirmation that orthogonality here is a consequence of the sine functions' symmetry, not of any particular box size. As a limiting-case check: setting $n=m$ instead (say both indices equal to 2) turns the same calculation into the normalization integral, giving $\int_0^L \psi_2^2\,dx = 1$ exactly (the $k=0$ term contributes $L$ instead of $0$, restoring the "$\delta_{mn}=1$" branch of the Kronecker delta rather than the "$=0$" branch checked above).

Example 2 — Applied: orthogonality of Legendre polynomials in the rigid-rotor / hydrogen-atom angular problem. The polar-angle part of the hydrogen 3d and 3s-type angular momentum eigenfunctions (and, more generally, any rigid-rotor eigenfunction with $m=0$) is built from Legendre polynomials $P_\ell(\cos\theta)$, eigenfunctions of the operator $\hat{L}^2$ with eigenvalue $\ell(\ell+1)\hbar^2$. Take the asymmetric, off-diagonal pair $\ell = 1$ and $\ell = 3$:

These belong to eigenvalues $2\hbar^2$ and $12\hbar^2$ respectively — manifestly distinct — so the theorem guarantees

Substitute $u = \cos\theta$, $du = -\sin\theta\,d\theta$, so the integral over $\theta \in [0,\pi]$ becomes an integral over $u \in [-1,1]$:

Integrate term by term:

So the bracket becomes $2 - 2 = 0$, and

confirming orthogonality by direct integration, not merely by appeal to the theorem. As a self-consistency check, the same substitution with $\ell=1$ paired against itself gives the normalization integral $\int_{-1}^1 u^2\,du = 2/3$, which after including the standard normalization constant $\sqrt{(2\ell+1)/2}$ for $P_1$ reproduces $\delta_{11}=1$ rather than $0$ — the expected other branch of the Kronecker delta.

Example 3 — Exam/competition level: using orthonormality to extract an LCAO expansion coefficient. Consider two 2p$_z$ atomic orbitals on adjacent carbon atoms in a conjugated $\pi$ system, forming (in the Hückel approximation) a bonding and antibonding pair of symmetry-adapted molecular orbitals:

where $\chi_A, \chi_B$ are normalized atomic 2p$_z$ orbitals and $S = \int \chi_A^*\chi_B\,d\tau$ is their overlap integral. Take the realistic, non-round value $S = 0.27$ for adjacent carbons at a typical bond distance. These two combinations are eigenfunctions of the molecular Hamiltonian (within the two-orbital model) belonging to distinct energies $E_+ = (\alpha+\beta)/(1+S)$ and $E_- = (\alpha-\beta)/(1-S)$ (the standard Hückel bonding/antibonding split), so orthogonality is guaranteed:

Suppose an experimentalist prepares the system (at $t=0$, immediately after excitation) in a non-stationary combination that happens to equal the unmixed orbital $\chi_A$ alone — a physically common initial condition right after a localized excitation event. We want the amplitude of the antibonding component in this state, i.e. $c_- = \int \psi_-^* \chi_A \, d\tau$. Write $\chi_A$ in terms of $\psi_+$ and $\psi_-$ by inverting the definitions:

so

By orthonormality of $\{\psi_+,\psi_-\}$ (each is normalized and they are mutually orthogonal, exactly as the theorem requires), the coefficient of $\psi_-$ read directly off this expansion is

Plug in $S = 0.27$:

The probability of measuring the system in the antibonding eigenstate is $|c_-|^2 = 0.6042^2 = 0.3650$, i.e. 36.5%, and by the analogous calculation $c_+ = \sqrt{2(1+S)}/2 = \sqrt{2.54}/2 = 0.7969$, giving $|c_+|^2 = 0.6350$, i.e. 63.5%. As a consistency check, $|c_+|^2 + |c_-|^2 = 0.6350 + 0.3650 = 1.0000$, confirming total probability is conserved — a direct payoff of using an orthonormal basis, since cross terms $c_+^*c_-\int\psi_+^*\psi_-\,d\tau$ vanish exactly by orthogonality and do not need to be tracked. As a limiting-case check, setting $S \to 0$ (orthogonal atomic orbitals, no overlap) gives $c_+ = c_- = \sqrt{2}/2 = 0.7071$, so $|c_+|^2=|c_-|^2=0.500$ — an even 50/50 split, exactly as expected when $\chi_A$ is symmetric with respect to the bonding/antibonding combinations in the no-overlap limit.

The orthogonality of eigenfunctions is not attributed to a single chemist; it emerges from the mathematics of self-adjoint (Hermitian) operators developed in the late 19th and early 20th centuries by mathematicians studying vibrating strings, heat conduction, and integral equations, and it was carried directly into quantum mechanics by the physicists who built the theory's mathematical foundations.