Two-Point Boundary-Value Problems

The two-point boundary-value problem is quietly everywhere a steady state or a confined mode is computed. In computational engineering, finite-element and finite-difference solvers for beams, membranes, and heat conduction reduce a BVP to a large but structurally identical linear system that carries the Fredholm alternative with it: a singular stiffness matrix flags a rigid-body mode or a buckling load exactly as the continuous determinant flags an eigenfunction. In quantum chemistry and photonics the same spectral problem sets molecular energy levels and the guided modes of an optical fibre, computed by shooting or relaxation with care taken over solvability at each trial eigenvalue. In control theory and orbital mechanics, two-point BVPs are the boundary conditions of Pontryagin's minimum principle — an optimal trajectory is fixed at both its initial and final states, and two-point solvers (multiple shooting, collocation) are the workhorses of spacecraft trajectory optimisation. The frontier keeps returning here too: adaptive meshes for singularly perturbed BVPs, spectral collocation for eigenvalue problems on unbounded domains, and the boundary conditions that select the radial Schrödinger and Sturm–Liouville solutions underlying every table of special functions.

Picture the two constants $c_1,c_2$ in the general solution as two dials on a control box. An initial-value problem sets both dials by reading the height and the slope of the curve at one place — two numbers, two dials, always possible and always unique. A boundary-value problem is stingier: it demands the curve pass through a peg at each end and says nothing about slope. Usually two dials can still hit two pegs. But if the family of curves happens to pivot about a fixed point — if every candidate is forced through the same node — then some peg heights are unreachable (no solution) while one privileged configuration lets the whole family swing free (infinitely many). That pivoting is exactly a vanishing solvability determinant.

Once the determinant is nonzero the operator has a genuine inverse, and that inverse has a name: the Green's function. It is the shape the system takes when you poke it with a unit point source at one location $\xi$, built to satisfy the boundary conditions automatically. Know the response to a single poke at every $\xi$ and linear superposition delivers the response to any distributed load $f$ — the integral $\int G(x,\xi)f(\xi)\,\mathrm{d}\xi$. For the prototypical Dirichlet problem $-u''=f$ on the unit interval this response is a simple tent, and its peak height is a number you can read straight off the geometry.

The eigenvalue connection is the same fact seen sideways. The operator $L$ fails to be invertible exactly when zero behaves like an eigenvalue — when a nontrivial homogeneous solution manages to satisfy both boundary conditions at once. That is resonance wearing a mathematician's coat: drive an undamped oscillator at one of its natural frequencies and the steady-state amplitude is undefined unless the forcing happens to be orthogonal to the resonant mode. The boundary-value problem, the Green's function, and the eigenvalue problem are three views of a single question: is $L$ invertible on the space its boundary conditions carve out?

The starting object is the most general regular linear second-order ODE on a finite interval. Two continuous coefficient functions and one forcing term define it; the interval's two endpoints are where the boundary data will live. Everything downstream is a statement about this equation together with conditions at $a$ and $b$.

where: $u(x)$ — the unknown function on the interval; $p,q$ — continuous coefficient functions; $f$ — the inhomogeneous forcing (source) term; $[a,b]$ — the finite interval whose two endpoints carry the boundary data. Reading it: a driven linear system whose free response spans a two-dimensional solution space. The boundary data most common in physics are separated: one linear relation among $u$ and $u'$ at each end. Tuning the coefficients selects Dirichlet, Neumann, or genuinely mixed (Robin) conditions, and setting the right-hand sides to zero gives the homogeneous conditions that define $L$'s domain.

where: $\alpha_i,\beta_i$ — constants selecting Dirichlet ($\alpha_2=0$), Neumann ($\alpha_1=0$), or mixed conditions; $\gamma_a,\gamma_b$ — prescribed boundary values; the homogeneous conditions take $\gamma_a=\gamma_b=0$. Reading it: each endpoint contributes exactly one scalar constraint on the two-parameter family. Now impose those two constraints on the general solution $u=u_p+c_1u_1+c_2u_2$. The result is a $2\times2$ linear system for $(c_1,c_2)$ whose coefficient matrix is the array of boundary functionals applied to the fundamental solutions. Its determinant is the single number that decides solvability.

where: $u_p$ — any particular solution of the forced equation (1); $u_1,u_2$ — a fundamental set of solutions of the homogeneous equation; $B_a,B_b$ — the boundary functionals at $a$ and $b$; $c_1,c_2$ — the constants the boundary conditions fix. Reading it: nonzero determinant means the boundary map from $(c_1,c_2)$ to boundary data is invertible. Recast in operator language, the whole trichotomy collapses to one crisp equivalence. Writing $L$ for the differential operator with the homogeneous boundary conditions built into its domain, unique solvability for every $f$ is equivalent to the homogeneous problem having only the trivial solution.

where: $L$ — the differential operator with boundary conditions folded into its domain; $\exists!$ — there exists a unique solution. Reading it: the operator is invertible iff its kernel is trivial, exactly as for a square matrix. In the invertible case the inverse is an integral operator with kernel $G(x,\xi)$. For a self-adjoint (Sturm–Liouville) operator $G$ has a compact closed form built from two homogeneous solutions, each adapted to one end, divided by their Wronskian and the leading coefficient.

where: $u_a$ — homogeneous solution satisfying the left boundary condition; $u_b$ — homogeneous solution satisfying the right boundary condition; $W(\xi)=u_a u_b' - u_a' u_b$ — their Wronskian; $p_0(x)$ — leading coefficient in self-adjoint form $(p_0u')'+\dots$, equal to $1$ for the plain $u''$ form. Reading it: $G$ is symmetric, $G(x,\xi)=G(\xi,x)$, the hallmark of a self-adjoint operator. With $G$ in hand the solution of the inhomogeneous problem is a single superposition integral: sum the responses to point sources of strength $f(\xi)$ across the interval. This is the continuous analogue of multiplying a load vector by an inverse matrix.

where: $G(x,\xi)$ — the response at $x$ to a unit point source at $\xi$; $f(\xi)$ — the distributed source. Reading it: linearity plus the point-source basis $\delta(x-\xi)$ gives every solution by integration. When the determinant vanishes, invertibility fails and a parameter enters. The homogeneous problem acquires nontrivial solutions only at special values $\lambda$ — the eigenvalues — with the nontrivial solutions as eigenfunctions against a positive weight $w$.

where: $\lambda$ — eigenvalue: the special parameter values admitting nontrivial $u$; $w(x)$ — positive weight function; $u$ — the eigenfunction. Reading it: the singular BVP is a spectral problem, the seed of Sturm–Liouville theory. At such a resonance the inhomogeneous problem is solvable only if the forcing is orthogonal to every homogeneous solution — the Fredholm condition, an $L^2$ inner-product statement.

where: $u_{\text{hom}}$ — the nontrivial homogeneous solution (kernel element) of the self-adjoint operator; $\langle\,,\,\rangle$ — the $L^2$ inner product on $[a,b]$. Reading it: forcing must have no overlap with the resonant mode, or no steady solution exists.

The character of a linear BVP $Lu=f$ with homogeneous boundary conditions falls into three mutually exclusive cases, decided entirely by the homogeneous problem $Lu=0$. The table is the Fredholm alternative sorted by outcome.

1 · Deflection of a uniformly loaded string (fully worked)

A string is stretched between $x=0$ and $x=1$ and carries a uniform transverse load, so its small displacement obeys $-u''=1$ with $u(0)=u(1)=0$. Find $u(x)$ using the Green's function, and locate and evaluate the maximum sag.

Step 1. The operator is $L=-\mathrm{d}^2/\mathrm{d}x^2$ with Dirichlet conditions, whose Green's function we built above: $G(x,\xi)=x(1-\xi)$ for $x\le\xi$ and $\xi(1-x)$ for $x\ge\xi$. Since the homogeneous problem $-u''=0$ with $u(0)=u(1)=0$ has only $u\equiv0$, the operator is invertible and a unique solution exists.

Step 2. Apply equation (6) with $f=1$: $u(x)=\int_0^1 G(x,\xi)\,\mathrm{d}\xi=(1-x)\!\int_0^x\xi\,\mathrm{d}\xi+x\!\int_x^1(1-\xi)\,\mathrm{d}\xi$. The two integrals are $(1-x)\tfrac{x^2}{2}$ and $x\tfrac{(1-x)^2}{2}$, which sum to $\tfrac{x(1-x)}{2}\big[x+(1-x)\big]$.

Step 3. The bracket collapses to $1$, giving the clean result $u(x)=\tfrac12 x(1-x)$. Check it directly: $u''=-1$, so $-u''=1=f$ ✓, and $u(0)=u(1)=0$ ✓.

Step 4. The maximum sits where $u'=\tfrac12(1-2x)=0$, i.e. $x=1/2$, by symmetry. Its value is $u(1/2)=\tfrac12\cdot\tfrac12\cdot\tfrac12=\mathbf{0.125}$. The string sags into a parabola $u=\tfrac12x(1-x)$ with maximum deflection $1/8=0.125$ at the midpoint.

2 · A Fredholm verdict: when does $u''+u=f$ on $[0,\pi]$ solve?

Decide the solvability of $u''+u=f$ with $u(0)=u(\pi)=0$ for (a) $f(x)=x$ and (b) $f(x)=\cos x$, and count the solutions in each case.

Step 1. The homogeneous problem has the nontrivial solution $u_{\text{hom}}=\sin x$ (we are at $\lambda=1$), so the regular theory does not apply and the deciding quantity is the overlap integral $\int_0^\pi f(x)\sin x\,\mathrm{d}x$.

Step 2. For $f=x$: integrate by parts, $\int_0^\pi x\sin x\,\mathrm{d}x=\big[\sin x-x\cos x\big]_0^\pi=(0-\pi\cos\pi)-(0)= \pi\approx3.14$. This is nonzero, so no solution exists — the constant-slope load is not orthogonal to the resonant mode.

Step 3. For $f=\cos x$: $\int_0^\pi\cos x\sin x\,\mathrm{d}x=\tfrac12\int_0^\pi\sin 2x\,\mathrm{d}x=\tfrac14\big[1-\cos2\pi\big]=0$. The orthogonality holds, so solutions exist.

Step 4. Because $\sin x$ lies in the kernel, whenever a particular solution exists the general solution is $u_p+c\sin x$ for arbitrary $c$. So $f=x$ gives no solution while $f=\cos x$ gives infinitely many — the same ODE, opposite verdicts, decided purely by orthogonality to $\sin x$.

3 · The particle in a box as a BVP turned eigenvalue problem

An electron is confined to a one-dimensional box of width $L=1\ \mathrm{nm}$ with $\psi(0)=\psi(L)=0$. Find the ground-state energy from the boundary-value spectrum.

Step 1. The stationary Schrödinger equation $-\tfrac{\hbar^2}{2m}\psi''=E\psi$ with the two Dirichlet conditions is the homogeneous BVP of equation (7). A nontrivial $\psi$ exists only when the solvability determinant vanishes, which happens at $E_n=\dfrac{n^2\pi^2\hbar^2}{2mL^2}$ with $\psi_n=\sqrt{2/L}\,\sin(n\pi x/L)$, $n=1,2,\dots$

Step 2. Insert $n=1$, $\hbar=1.055\times10^{-34}\ \mathrm{J\,s}$, $m=9.11\times10^{-31}\ \mathrm{kg}$, $L=10^{-9}\ \mathrm{m}$. The numerator is $\pi^2\hbar^2=9.870\times(1.113\times10^{-68})=1.098\times10^{-67}$; the denominator is $2mL^2=1.822\times10^{-48}$.

Step 3. Divide: $E_1=1.098\times10^{-67}/1.822\times10^{-48}=6.02\times10^{-20}\ \mathrm{J}$, and dividing by $1.602\times10^{-19}\ \mathrm{J/eV}$ gives $E_1=\mathbf{0.376\ eV}$. The first excited level is $E_2=4E_1=1.50\ \mathrm{eV}$. The discreteness of the box spectrum is nothing but the solvability determinant vanishing only at special energies — the bridge from BVPs to quantisation.

The two-point boundary-value problem is where elementary ODE technique meets the spectral theory that runs modern physics, and the hinge is a single dichotomy first proved not for differential equations at all but for integral equations — by a Swedish actuary-turned-mathematician whose name the alternative now carries.

Texts. Stone & Goldbart, Mathematics for Physics, ch. 1–3 (calculus of variations, ODEs, Green functions) · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., ch. 7–8 (ODEs and Sturm–Liouville theory) · Hassani, Mathematical Physics, ch. 20–22 (Green's functions and ODEs) · Riley, Hobson & Bence, Mathematical Methods for Physics and Engineering, 3rd ed., ch. 15–17 · Byron & Fuller, Mathematics of Classical and Quantum Physics, ch. 6 (Green's functions) · Courant & Hilbert, Methods of Mathematical Physics, vol. I, ch. V–VI · Coddington & Levinson, Theory of Ordinary Differential Equations, ch. 7 (boundary-value problems). Key anchor results collected here for reference: the $2\times2$ solvability determinant criterion follows from Cramer's rule; the Green's function for $-u''$ with Dirichlet conditions on $[0,1]$ is $G(x,\xi)=x_<(1-x_>)$ with $x_<=\min(x,\xi)$ and $x_>=\max(x,\xi)$; the uniform-load deflection is $u=\tfrac12x(1-x)$ with maximum $1/8=0.125$; the Fredholm integrals evaluate to $\int_0^\pi x\sin x\,\mathrm{d}x=\pi$, $\int_0^\pi\cos x\sin x\,\mathrm{d}x=0$, and $\int_0^\pi\sin^2x\,\mathrm{d}x=\pi/2$; and the ground-state particle-in-a-box energy is $E_1=\pi^2\hbar^2/2mL^2=0.376\ \mathrm{eV}$ for $L=1\ \mathrm{nm}$.

Prerequisites. Linear Second-Order ODEs · The Wronskian and Linear Independence · Existence and Uniqueness Theorems for ODEs · Variation of Parameters. Leads to. Sturm–Liouville Theory · The Method of Frobenius · Ordinary and Singular Points.

Ask the tutor: ask me to walk through the solvability determinant, build a Green's function for your own boundary conditions, or show exactly when your forcing tips a resonant BVP from no solution to infinitely many.