Self-Adjointness and Boundary Conditions

The abstract theory has become laboratory hardware. Ultracold atoms are trapped in ring-shaped optical potentials where synthetic gauge fields impose exactly the twisted condition $\psi(L)=e^{\mathrm{i}\theta}\psi(0)$ of Worked Example 2 — self-adjoint extensions realized on demand, with persistent currents that jump as $\theta$ threads a flux quantum. Box traps with sharp optical walls realize the Dirichlet well of Worked Example 1 for interacting gases. Quantum graphs — networks of one-dimensional wires meeting at vertices — model carbon nanotubes, photonic circuits, and mesoscopic devices, and the physics of each vertex is the self-adjoint extension chosen there. The old "fall to the center" problem of the $-c/r^2$ potential is a limit-circle endpoint requiring an extension, whose parameter resurfaces in near-horizon black-hole physics and in the Efimov effect, where a discrete scaling symmetry is the signature of a missing boundary condition. Even numerical spectral methods lean on it: a discretized operator inherits real eigenvalues only if the boundary conditions preserve symmetry — getting that wrong is a classic source of spurious complex modes.

Think of functions as vectors in the infinite-dimensional space $L^2(a,b)$. A finite Hermitian matrix automatically has real eigenvalues and orthogonal eigenvectors — no fine print. We want a differential operator to inherit the same guarantees, but in infinite dimensions the "matrix" has edges: integration by parts, the continuous analogue of transposing, spills a boundary term at the endpoints that a finite matrix never produces. Self-adjointness is the discipline of choosing boundary conditions so this edge term vanishes for every pair at once — and, subtly, so the domain is neither too small nor too large.

Picture the concomitant $Q(u,v)=p(\bar u'v-\bar uv')$ as a probability flux leaking across each wall. Symmetry says no net flux escapes: Dirichlet ($u=0$), Neumann ($u'=0$), Robin ($u'=\alpha u$) and periodic conditions are all valves that jam the flux to zero. The infinite-dimensional surprise is that closing the valve for the chosen domain is not enough — the adjoint quietly claims the largest domain the identity permits, so $\mathcal{D}(L^\dagger)$ can exceed $\mathcal{D}(L)$. A self-adjoint operator is "just big enough": constrained enough to be symmetric, loose enough that its adjoint has no extra room. The payoff is a real, orthogonal, complete spectrum — as concrete as the energy ladder of a particle trapped between two walls.

The most general formally self-adjoint second-order operator is the Sturm–Liouville form — not an assumption but a theorem: any regular second-order operator can be multiplied by an integrating factor until its symbol is self-adjoint with respect to a weight $w$. Everything downstream — reality, orthogonality, completeness — flows from this template.

where: $p(x)$ — coefficient of the highest derivative, $p>0$ on the open interval; $p\to0$ marks a singular endpoint; $q(x)$ — real potential / zeroth-order coefficient; $w(x)$ — positive weight fixing the inner product $\langle u,v\rangle=\int w\,\bar u v\,\mathrm{d}x$; $u(x)$ — function in $\mathcal{D}(L)\subset L^2(a,b;w)$. Reading: $p$, $q$, $w$ are the DNA of the problem; $p$ vanishing at an end is the flag for a singular endpoint. Lagrange's identity is the algebraic heart. For the unweighted part $L_0=-(pu')'+qu$ the antisymmetric combination collapses to a total derivative, which is what makes the boundary term a single endpoint evaluation rather than an integral.

where: $L_0$ — unweighted formally self-adjoint part, $L_0=wL$; $p(uv'-u'v)$ — the bilinear concomitant $Q(u,v)$, a $p$-weighted Wronskian. Reading: the antisymmetric combination is a pure derivative, so integrating it is free. Integrating (2) and conjugating gives Green's boundary formula. The difference between $\langle L_0u,v\rangle$ and $\langle u,L_0v\rangle$ is entirely the concomitant evaluated at the two ends; symmetry is the demand that it vanish.

where: $\langle\cdot,\cdot\rangle$ — the $L^2$ inner product $\int_a^b\bar f g\,\mathrm{d}x$, conjugate-linear in the first slot; $Q(u,v)|_a^b$ — boundary term, required to vanish for all $u,v\in\mathcal{D}(L)$. Reading: no interior term survives — the entire obstruction to symmetry lives on the boundary. Separated (Robin) conditions shut the valve at each regular endpoint independently: a single homogeneous real relation between $u$ and $u'$ at each end forces the concomitant to vanish there.

where: $\alpha_i,\beta_i$ — real constants with $(\alpha_1,\alpha_2)\neq(0,0)$ and $(\beta_1,\beta_2)\neq(0,0)$. Reading: Dirichlet ($\alpha_2=0$), Neumann ($\alpha_1=0$) and Robin (both nonzero) are all special cases; each kills $Q$ at its own endpoint. Self-adjointness pays off in the spectrum. Feeding two eigenfunctions into Green's formula, with the boundary term gone, yields both spectral cornerstones at once: real eigenvalues and weighted-orthogonal eigenfunctions.

where: $\lambda_n$ — eigenvalue of $Lu_n=\lambda_n u_n$ (equivalently $L_0u_n=\lambda_n w u_n$); $\langle u_n,u_m\rangle_w=\int_a^b w\,\bar u_n u_m\,\mathrm{d}x$ — weighted overlap. Reading: put $m=n$ to force $\lambda$ real; put $\lambda_n\neq\lambda_m$ to force orthogonality — under the weight $w$, never the bare measure. When an operator refuses to be self-adjoint, von Neumann's deficiency indices quantify the failure by counting the square-integrable solutions of $L^\dagger u=\pm\mathrm{i}u$ — the adjoint's eigenfunctions at the non-real test values $\pm\mathrm{i}$.

where: $n_+,n_-$ — number of $L^2$ solutions of $L^\dagger u=\pm\mathrm{i}u$; $L^\dagger$ — the Hilbert-space adjoint on its maximal domain. Reading: a self-adjoint operator can have no eigenfunction at $\pm\mathrm{i}$, so any that survive measure the shortfall. The indices decide everything about extensions: equal indices mean the operator can be completed to self-adjoint, unequal indices mean it cannot be, whatever boundary condition you try.

where: $n_+=n_-=0$ — $L$ already essentially self-adjoint (unique closure); $n_+=n_-=n>0$ — a $U(n)$ family of distinct extensions; $n_+\neq n_-$ — no self-adjoint realisation at all. Reading: the count is a verdict — a unique answer, a $U(n)$ menu, or no answer.

How the self-adjointness requirement specializes across the regimes that occur in practice:

1 · The infinite square well as a self-adjoint problem (fully worked)

An electron is confined to a one-dimensional box of width $L=1$ nm with impenetrable walls. Treat $-\mathrm{d}^2/\mathrm{d}x^2$ as a Sturm–Liouville operator ($p=w=1$, $q=0$) on $[0,L]$ with Dirichlet conditions $\psi(0)=\psi(L)=0$. Verify self-adjointness, find the spectrum, and confirm orthonormality and reality with numbers.

Step 1. Check the boundary term. Green's formula gives $Q(\psi,\phi)|_0^L=[\bar\psi'\phi-\bar\psi\phi']_0^L$; Dirichlet conditions set $\psi=\phi=0$ at both ends, so every term vanishes. Both endpoints are regular with indices $(2,2)$, and the two Dirichlet conditions are exactly the constraints needed, so the operator is genuinely self-adjoint — no extra room in $\mathcal{D}(L^\dagger)$.

Step 2. Solve the eigenproblem. $-\psi''=\lambda\psi$ with $\psi(0)=\psi(L)=0$ forces $\psi_n(x)=\sqrt{2/L}\,\sin(n\pi x/L)$ and $\lambda_n=(n\pi/L)^2$ for $n=1,2,3,\dots$ — a real, discrete spectrum, as self-adjointness guarantees. The energies follow from $E_n=\hbar^2\lambda_n/2m=n^2\pi^2\hbar^2/2mL^2$.

Step 3. Put in numbers. With $\hbar=1.055\times10^{-34}$ J·s, $m=9.109\times10^{-31}$ kg, $L=10^{-9}$ m, the ground state is $E_1=\pi^2\hbar^2/2mL^2=6.02\times10^{-20}$ J $=\mathbf{0.376\ eV}$. The ladder scales as $n^2$: $E_2=4E_1=\mathbf{1.504\ eV}$ and $E_3=9E_1=\mathbf{3.384\ eV}$.

Step 4. Confirm orthonormality. The weighted overlap (here $w=1$) is $\int_0^L\frac{2}{L}\sin\frac{n\pi x}{L}\sin\frac{m\pi x}{L}\,\mathrm{d}x=\delta_{nm}$; numerically $\int_0^1 2\sin(n\pi x)\sin(m\pi x)\,\mathrm{d}x$ evaluates to $1.000$ for $n=m$ and $0.000$ for $n\neq m$. The spectrum is real and the eigenfunctions orthonormal — exactly the guarantees self-adjointness buys, made concrete: E₁ = 0.376 eV, Eₙ = n²E₁, ⟨ψₙ,ψₘ⟩ = δₙₘ.

2 · Momentum on a ring and the twist phase

A particle lives on a ring of circumference $L$, i.e. $-\mathrm{i}\,\mathrm{d}/\mathrm{d}x$ on $[0,L]$ with the endpoints identified. Show the self-adjoint realisations form a one-parameter family and find their spectra.

Step 1. From Derivation 3, both deficiency solutions $e^{\pm x}$ are square-integrable on the bounded interval, so $n_+=n_-=1$: the extensions form a $U(1)$ family parametrised by a phase $\theta\in[0,2\pi)$ through $\psi(L)=e^{\mathrm{i}\theta}\psi(0)$.

Step 2. For each $\theta$ the eigenfunctions are $\psi_n(x)=L^{-1/2}e^{\mathrm{i}k_n x}$ with $k_nL=2\pi n+\theta$, so $k_n=(2\pi n+\theta)/L$ and the real eigenvalues are $\hbar k_n$. The concomitant $-\mathrm{i}\,\bar\psi\phi|_0^L$ vanishes because both functions carry the same phase, confirming self-adjointness for every $\theta$.

Step 3. Interpret the phase. $\theta=0$ is ordinary periodicity; nonzero $\theta$ is an Aharonov–Bohm flux $\Phi=\theta\,\hbar/e$ threading the ring, shifting the entire ladder by $\theta/L$ without changing its spacing $2\pi/L$. Different self-adjoint extensions are different fluxes — the "choice of boundary condition" is a measurable magnetic flux, not a convention.

Self-adjointness sits where nineteenth-century boundary-value problems meet twentieth-century functional analysis. The name most bound to its subtlest corner — the singular endpoint, where $p$ vanishes and ordinary boundary conditions lose their meaning — is Hermann Weyl.

Texts. Stone & Goldbart, Mathematics for Physics, Ch. 1–4 (function spaces, linear differential operators, the bilinear concomitant, self-adjointness) · Hassani, Mathematical Physics, 2nd ed., Ch. 19 (Sturm–Liouville systems) · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., §8.2 (Hermitian operators, Sturm–Liouville theory) · Byron & Fuller, Mathematics of Classical and Quantum Physics, Ch. 5 · Reed & Simon, Methods of Modern Mathematical Physics, Vol. II, Ch. X (self-adjointness, deficiency indices) · Nakahara, Geometry, Topology and Physics, Ch. 12 (index theorems and boundary conditions).

Provenance. Lagrange's identity and Green's formula are cross-checked against Stone–Goldbart Ch. 4 and Arfken–Weber–Harris §8.2; the reality/orthogonality derivation against Hassani Ch. 19 and Byron–Fuller Ch. 5; the deficiency-index computation for $-\mathrm{i}\,\mathrm{d}/\mathrm{d}x$ against Reed–Simon Vol. II §X.1. A verifier should confirm: (1) the cancellation of $q$ and of the $pu'v'$ cross-terms leaving the concomitant $p(uv'-u'v)$; (2) the box spectrum $E_n=n^2\pi^2\hbar^2/2mL^2$ with $E_1=0.376$ eV, $E_2=1.504$ eV, $E_3=3.384$ eV for an electron in $L=1$ nm; (3) orthonormality $\int_0^1 2\sin(n\pi x)\sin(m\pi x)\,\mathrm{d}x=\delta_{nm}$; (4) deficiency indices $(1,0)$, $(0,0)$, $(1,1)$ for $-\mathrm{i}\,\mathrm{d}/\mathrm{d}x$ on half-line, line, and interval. No fabricated constants are used.

Prerequisites. Linear Second-Order ODEs · The Wronskian and Linear Independence · Two-Point Boundary-Value Problems · Ordinary and Singular Points. Leads to. Two-Point Boundary-Value Problems · The Method of Frobenius.

Ask the tutor: Ask me to walk through Weyl's limit-point/limit-circle test on a real singular equation — Bessel, Legendre, or the hydrogen radial operator — and watch the boundary condition appear (or refuse to) on its own.