Linear Second-Order ODEs

The structure theory of linear second-order ODEs is the daily working grammar of computational and experimental physics. Every shooting-method solver for the time-independent Schrödinger equation $-\tfrac{\hbar^2}{2m}\psi''+V(x)\psi=E\psi$ leans on the two-dimensional solution space to match wavefunctions across a potential step and on the Wronskian to detect dependence when hunting bound-state energies. In gravitational-wave astronomy, the ringdown of a perturbed black hole obeys the Regge–Wheeler and Teukolsky equations — linear second-order ODEs whose quasinormal-mode spectrum LIGO and Virgo now measure directly, the effective potential setting damped-oscillation frequencies exactly as in the first figure. Control engineers still tune servo loops and seismometers to critical damping, the repeated-root case, for the fastest settling without overshoot. And the special functions born from separating the wave and Laplace equations — Bessel functions in an optical fibre, Legendre in multipole radiation, Hermite in the quantum oscillator — are variable-coefficient members of this family, handled by the power-series and Frobenius methods this structure theory makes rigorous. An open frontier is the theory of ODEs with irregular singular points, where resurgence and exact WKB methods are reshaping how physicists resum divergent series.

Treat $L=\dfrac{\mathrm{d}^2}{\mathrm{d}x^2}+p\dfrac{\mathrm{d}}{\mathrm{d}x}+q$ as a linear machine acting on functions. Asking for $L[y]=0$ is asking for its kernel; asking for $L[y]=f$ is asking to invert it on the forcing $f$. The kernel is two-dimensional because pinning down $y$ and $y'$ at a single point fixes the entire solution — two numbers, two dimensions — and physically those two numbers are the position and velocity a Newtonian particle needs to start. Every second-order equation is $F=ma$ in disguise, and the freedom to choose where and how fast you start is the two-parameter family of solutions. The clearest laboratory for this is the constant-coefficient oscillator $\ddot{x}+b\dot{x}+cx=0$, where the sign of the discriminant $\Delta=b^2-4c$ sorts every possible behaviour into three regimes.

The split $y=y_c+y_p$ is the physics of a driven system made visible. The complementary part $y_c$ is the transient — the free oscillation carrying the initial data, doomed to decay whenever there is damping — and the particular part $y_p$ is the steady state, the response locked in step with the driving force. Linearity is exactly the licence to add them. Watch a lightly damped oscillator driven periodically: the transient blooms and dies while the steady state marches on, and after a few damping times only $y_p$ remains — which is why an instrument settles into a pure tone.

Finally, the Wronskian is the quantitative meter for "genuinely different solutions". Reading each solution as a phase-space vector $(y_i,y_i')$, the determinant $W=y_1y_2'-y_2y_1'$ is the signed area of the parallelogram they span. Two solutions are independent exactly when that area is nonzero, and Abel's identity guarantees the area can never sneak through zero at an interior point without being zero everywhere.

Everything begins by clearing the leading coefficient so the highest derivative stands alone. The result is the standard form, the shape in which the coefficient functions $p$ and $q$ are read off and the entire structure theory applies.

where: $y(x)$ — the unknown function to be determined; $p(x),q(x)$ — coefficient functions, assumed continuous on the interval of interest; $f(x)$ — the forcing or inhomogeneous term, set to zero for the homogeneous equation. Reading it: the equation is homogeneous when $f\equiv0$ and inhomogeneous otherwise, and you must divide out any leading coefficient $a_2(x)$ before trusting $p=a_1/a_2$ and $q=a_0/a_2$. The reason this class is tractable is a single algebraic property of the operator $L=\tfrac{\mathrm{d}^2}{\mathrm{d}x^2}+p\tfrac{\mathrm{d}}{\mathrm{d}x}+q$. Because differentiation is linear, so is $L$, and superposition follows immediately.

where: $L$ — the differential operator; $\alpha,\beta$ — arbitrary constants; $y_1,y_2$ — any two functions in the domain. Reading it: a sum of homogeneous solutions is again a homogeneous solution, which is precisely what makes the homogeneous solution set a vector space rather than a mere collection of curves. That vector space is two-dimensional, so a full solution of the inhomogeneous problem carries two arbitrary constants plus one fixed particular piece.

where: $y_1,y_2$ — linearly independent homogeneous solutions (a fundamental set); $y_p$ — any single particular solution of the inhomogeneous equation; $c_1,c_2$ — constants fixed by initial or boundary conditions. Reading it: the complementary combination $c_1y_1+c_2y_2$ sweeps out the whole kernel, and adding $y_p$ shifts that plane without changing its dimension. To test whether a candidate pair actually is a fundamental set, form the Wronskian determinant.

where: $W(x)$ — the Wronskian; a value $W\ne0$ certifies linear independence and hence that $\{y_1,y_2\}$ spans the solution space. Reading it: $W$ is the phase-space area spanned by the two solution vectors $(y_1,y_1')$ and $(y_2,y_2')$. The Wronskian is not just a pointwise test — it satisfies its own first-order equation driven only by $p$, whose solution is Abel's identity.

where: $x_0$ — a reference point in the interval; $p(t)$ — the first-derivative coefficient in standard form. Reading it: because the exponential never vanishes, $W$ is either identically zero or nowhere zero — independence is a property of the whole interval, decided at any one point. If one nontrivial solution $y_1$ is already known, a second independent one can be manufactured directly by reduction of order.

where: $y_1$ — one known nontrivial solution; $y_2$ — the second, independent solution built from it. Reading it: the numerator $e^{-\int p}$ is the Wronskian up to a constant, so this formula is Abel's identity solved for the missing partner. Once a fundamental set is in hand, a particular solution for any forcing $f$ comes from variation of parameters — no guessing required.

where: $W$ — the Wronskian of $y_1,y_2$; $f$ — the forcing term in standard form. Reading it: this is Cramer's rule applied to the promoted constants, and it works for arbitrary continuous $f$, unlike undetermined coefficients. When $p,q$ are constant the whole apparatus collapses to algebra: substitute $y=e^{rx}$ and the equation becomes a polynomial in $r$.

where: $b,c$ — the constant coefficients of $y''+by'+cy=0$; $r$ — the exponent in the trial solution $y=e^{rx}$. Reading it: the discriminant $b^2-4c$ decides between two real exponentials, one repeated exponential, or a decaying sinusoid — the three regimes of the first figure.

The constant-coefficient equation $y''+by'+cy=0$, with roots $r_{1,2}=\tfrac{-b\pm\sqrt{b^2-4c}}{2}$, already displays every qualitative behaviour a linear second-order ODE can have. The three homogeneous regimes plus the resonant forced case are collected below.

1 · A constant-coefficient initial-value problem (confidence builder)

Solve $y''-5y'+6y=0$ with $y(0)=1$ and $y'(0)=0$, and confirm the two building-block solutions are independent.

Step 1. Substitute $y=e^{rx}$ to get the characteristic equation $r^2-5r+6=0$, which factors as $(r-2)(r-3)=0$, so $r=2$ and $r=3$. Two distinct real roots put us in the overdamped regime, and the general solution is $y=c_1e^{2x}+c_2e^{3x}$.

Step 2. Check independence with the Wronskian: $W=y_1y_2'-y_2y_1'=e^{2x}(3e^{3x})-e^{3x}(2e^{2x})=3e^{5x}-2e^{5x}=e^{5x}$. Since $e^{5x}\ne0$ everywhere, $\{e^{2x},e^{3x}\}$ is a fundamental set. Notice this matches Abel's identity: with $p=-5$, equation (5) predicts $W\propto e^{-\int(-5)\,\mathrm{d}x}=e^{5x}$, exactly what we computed.

Step 3. Impose the initial data. From $y(0)=c_1+c_2=1$ and $y'(0)=2c_1+3c_2=0$, subtract twice the first from the second to get $c_2=-2$, then $c_1=3$. The unique solution is $y=3e^{2x}-2e^{3x}$, and a quick check confirms $y(0)=3-2=1$ and $y'(0)=6-6=0$ as required.

2 · Resonance and the secular term (fully worked)

An undamped oscillator is driven exactly at its natural frequency: $\ddot{x}+4x=\cos(2t)$, starting from rest, $x(0)=0,\ \dot{x}(0)=0$. Find $x(t)$ and describe the long-time behaviour.

Step 1. The homogeneous equation $\ddot{x}+4x=0$ has characteristic roots $r=\pm2i$ (natural frequency $\omega_0=2$), so the complementary solution is $x_c=c_1\cos2t+c_2\sin2t$. The drive frequency $\omega=2$ coincides with $\omega_0$, so a plain $\cos2t$ trial would be swallowed by $x_c$ — the resonant flag.

Step 2. Because the forcing sits on a natural frequency, the trial solution must carry an extra factor of $t$: take $x_p=At\sin2t$. Then $\dot{x}_p=A\sin2t+2At\cos2t$ and $\ddot{x}_p=4A\cos2t-4At\sin2t$, so $\ddot{x}_p+4x_p=4A\cos2t$. Matching to $\cos2t$ gives $4A=1$, hence $A=\tfrac14$ and $x_p=\tfrac{t}{4}\sin2t$.

Step 3. Apply the initial conditions to the full solution $x=\tfrac{t}{4}\sin2t+c_1\cos2t+c_2\sin2t$. At $t=0$: $x(0)=c_1=0$. Differentiate and evaluate: $\dot{x}(0)=2c_2=0$, so $c_2=0$. The motion is $x(t)=\tfrac{t}{4}\sin2t$ — a sinusoid inside a linearly growing envelope $\pm t/4$: at $t=10\,$s the amplitude has reached $2.5$, at $t=40\,$s it is $10$. With no damping to cap it, the response climbs without bound — the mathematical face of undamped resonance, and why structures are engineered to keep drive frequencies off their natural modes.

The theory of linear second-order ODEs is an eighteenth-century collective achievement — but its decisive computational key, the exponential ansatz that turns calculus into algebra, belongs to Leonhard Euler.

Texts. Stone & Goldbart, Mathematics for Physics, Ch. 1 and Ch. 3 (linear ODEs) · Arfken & Weber, Mathematical Methods for Physicists, 7th ed., Ch. 7–8 · Hassani, Mathematical Physics, Ch. 25–26 (ODEs and their solutions) · Riley, Hobson & Bence, Mathematical Methods for Physics and Engineering, 3rd ed., Ch. 15–16 · Byron & Fuller, Mathematics of Classical and Quantum Physics, Ch. 2 · Boyce & DiPrima, Elementary Differential Equations and Boundary Value Problems, Ch. 3 · Coddington & Levinson, Theory of Ordinary Differential Equations, Ch. 3. Key results to keep in mind: $\dim\ker L=2$ follows from the companion first-order system and existence-uniqueness; Abel's identity satisfies $W'=-pW$; the variation-of-parameters formulas follow from Cramer's rule with determinant $W\ne0$; the repeated-root case gives $y_2=xe^{rx}$ via reduction of order; the worked IVP yields $r^2-5r+6=0\Rightarrow r=2,3$, $W=e^{5x}$, $y=3e^{2x}-2e^{3x}$; and the resonant secular solution $x_p=\tfrac{t}{4}\sin2t$ satisfies $\ddot{x}+4x=\cos2t$.

Prerequisites. First-Order Ordinary Differential Equations · Existence and Uniqueness Theorems for ODEs · Integrating Factors and Exact Equations. Leads to. The Wronskian and Linear Independence · Reduction of Order · Variation of Parameters · Power Series Solutions · The Method of Frobenius · Sturm–Liouville Theory.

Ask the tutor: Give me any linear second-order ODE — constant or variable coefficient, homogeneous or driven — and I will walk you from standard form through the Wronskian, the fundamental set, and a particular solution, spelling out why each step is forced.