Second-Order Equations

⚡ Key Takeaways

A second-order (ordinary) differential equation relates an unknown function $y(x)$ to its first and second derivatives. The central case — the linear second-order ODE $a(x)y'' + b(x)y' + c(x)y = f(x)$ — governs oscillations, waves, diffusion, and quantum systems. Key result: its general solution is the sum of any particular solution plus a two-parameter family of homogeneous solutions spanned by two linearly independent functions. Appears everywhere in physics, engineering, and applied math; theory matured through Euler, d'Alembert, and Sturm–Liouville (18th–19th centuries).

💡 The Core Insight

A linear second-order equation behaves like a linear map on a space of functions. The homogeneous equation $L[y]=0$ has a solution set that is a two-dimensional vector space; the inhomogeneous equation $L[y]=f$ has solutions forming an affine copy of that space. Therefore: find two independent homogeneous solutions, find one particular solution, and you have them all. The number two is not arbitrary — it equals the order, which equals the number of initial conditions ( $y$ and $y'$ ) needed to determine motion uniquely.

DefinitionGeneral Second-Order ODE

A second-order ordinary differential equation is any relation

1

\quad F\big(x,\, y,\, y',\, y''\big) = 0

Derivation hint: this is simply the definition — "second-order" means $y''$ is the highest derivative present and $y''$ actually appears (so $\partial F/\partial y'' \neq 0$ ). Here $x$ is the independent variable, $y=y(x)$ the unknown function, $y'=dy/dx$ , and $y''=d^2y/dx^2$ . The equation is explicit if it can be solved for the top derivative as $y'' = G(x,y,y')$ .

DefinitionLinear Second-Order ODE

The equation is linear if $F$ is a first-degree polynomial in $y, y', y''$ :

2

\quad a_2(x)\,y'' + a_1(x)\,y' + a_0(x)\,y = f(x)

Derivation hint: linearity means no products like $yy'$ , no powers $(y')^2$ , no transcendental functions of $y$ — only $y,y',y''$ each multiplied by functions of $x$ alone. Symbols: $a_2,a_1,a_0$ are the coefficient functions; $f(x)$ is the forcing (or source) term. If $f\equiv 0$ the equation is homogeneous; otherwise inhomogeneous. Dividing by $a_2$ (where $a_2\neq 0$ ) gives the standard form

3

\quad y'' + p(x)\,y' + q(x)\,y = g(x), \qquad p=\frac{a_1}{a_2},\; q=\frac{a_0}{a_2},\; g=\frac{f}{a_2}

Derivation hint: divide [2] through by $a_2(x)$ , valid on any interval where $a_2$ does not vanish; points where $a_2=0$ are called singular points. Define the linear differential operator $L[y] := y'' + p(x)y' + q(x)y$ , so the equation reads $L[y]=g$ .

TheoremExistence and Uniqueness

Let $p, q, g$ be continuous on an open interval $I$ containing $x_0$ . Then for any numbers $y_0, y_0'$ the initial value problem

4

\quad y'' + p(x)y' + q(x)y = g(x), \quad y(x_0)=y_0,\; y'(x_0)=y_0'

Derivation hint: convert to a first-order system $\mathbf u' = A(x)\mathbf u + \mathbf b$ with $\mathbf u=(y,y')^\top$ and apply the Picard–Lindelöf theorem; continuity of $p,q,g$ gives a Lipschitz condition. has a unique solution on the entire interval $I$ .

TheoremStructure of the Solution Set

The solutions of the homogeneous equation $L[y]=0$ form a two-dimensional vector space. If $y_1, y_2$ are linearly independent solutions (a fundamental set), the general homogeneous solution is

5

\quad y_h(x) = c_1 y_1(x) + c_2 y_2(x), \qquad c_1,c_2 \in \mathbb{R}

Derivation hint: superposition holds because $L$ is linear, so $L[c_1y_1+c_2y_2]=c_1L[y_1]+c_2L[y_2]=0$ ; dimension two follows from matching the two initial data $y(x_0),y'(x_0)$ uniquely (Theorem [4]). The general solution of the inhomogeneous equation $L[y]=g$ is

6

\quad y(x) = y_p(x) + c_1 y_1(x) + c_2 y_2(x)

Derivation hint: if $y$ and $y_p$ both satisfy $L[\cdot]=g$ , then $L[y-y_p]=g-g=0$ , so $y-y_p$ is homogeneous; rearrange. where $y_p$ is any particular solution.

DefinitionWronskian and Linear Independence

The Wronskian of two solutions is the determinant

7

\quad W(y_1,y_2)(x) = \begin{vmatrix} y_1 & y_2 \\ y_1' & y_2' \end{vmatrix} = y_1 y_2' - y_2 y_1'

Derivation hint: it is the determinant of the coefficient matrix obtained when solving for $c_1,c_2$ from $y(x_0)=y_0$ , $y'(x_0)=y_0'$ . Two solutions are linearly independent on $I$ iff $W\neq 0$ somewhere on $I$ . The Wronskian obeys Abel's identity:

8

\quad W(x) = W(x_0)\,\exp\!\left(-\int_{x_0}^{x} p(t)\,dt\right)

Derivation hint: differentiate [7], substitute $y_i'' = -p y_i' - q y_i$ ; the $q$ -terms cancel, leaving $W' = -p\,W$ , a separable first-order ODE. A consequence: the Wronskian is either identically zero or never zero on $I$ .

ProofThe General Solution Has Exactly Two Free Constants

Step 1 — Reduce to a first-order system. Set $u_1 = y$ , $u_2 = y'$ . Then

9

\quad \mathbf{u}' = \begin{pmatrix} 0 & 1 \\ -q(x) & -p(x) \end{pmatrix}\mathbf{u} + \begin{pmatrix} 0 \\ g(x) \end{pmatrix}, \qquad \mathbf u = \begin{pmatrix} u_1 \\ u_2\end{pmatrix}

Derivation hint: $u_1' = y' = u_2$ trivially, and $u_2' = y'' = -p y' - q y + g = -p u_2 - q u_1 + g$ from the standard form [3].

Step 2 — Apply existence/uniqueness to the system. With $p,q,g$ continuous, the matrix entries are continuous, so by the vector Picard–Lindelöf theorem a unique solution exists for each initial vector $\mathbf u(x_0)=(y_0,y_0')^\top$ .

Step 3 — Count dimensions. The map (solution) $\mapsto$ (its value $\mathbf u(x_0)$ ) is a linear isomorphism from the homogeneous solution space onto $\mathbb{R}^2$ : it is linear (superposition), injective (uniqueness — same data forces same solution), and surjective (existence — every datum is achieved). An isomorphism preserves dimension, so the solution space has dimension exactly $\boxed{2}$ .

Step 4 — Build a basis. Choose solutions $y_1, y_2$ with initial data $(1,0)$ and $(0,1)$ respectively. Their Wronskian at $x_0$ is $\det I = 1 \neq 0$ , so they are independent and span. ∎

∎

ProofVariation of Parameters Gives a Particular Solution

Suppose $y_1,y_2$ solve the homogeneous equation. Seek $y_p = v_1 y_1 + v_2 y_2$ with functions $v_1,v_2$ . Impose the convenient constraint $v_1' y_1 + v_2' y_2 = 0$ . Then computing $y_p''$ and substituting into $L[y_p]=g$ yields the linear system

10

\quad \begin{cases} v_1' y_1 + v_2' y_2 = 0 \\[2pt] v_1' y_1' + v_2' y_2' = g \end{cases}

Derivation hint: differentiate $y_p$ once using the constraint to kill the $v_i'$ terms, differentiate again, then insert into $L[y_p]$ ; all $v_i$ (un-primed) terms vanish because $y_i$ solve $L[y_i]=0$ . Solving by Cramer's rule (denominator $=W$ ):

11

\quad v_1' = -\frac{y_2\,g}{W}, \qquad v_2' = \frac{y_1\,g}{W}, \qquad y_p = -y_1\!\int\!\frac{y_2 g}{W}dx + y_2\!\int\!\frac{y_1 g}{W}dx

Derivation hint: invert the $2\times2$ system [10]; the determinant is precisely the Wronskian [7], which is nonzero for independent solutions. ∎

∎

mathematician · Charles-François Sturm | 1803–1855

The Swiss-French mathematician Charles Sturm, working with Joseph Liouville in the 1830s, transformed the study of second-order linear equations. Rather than seeking closed-form solutions, Sturm asked qualitative questions: How many zeros does a solution have? How do solutions of two related equations interlace? His comparison and oscillation theorems showed that the coefficient $q(x)$ controls how rapidly solutions oscillate, foreshadowing modern spectral theory. The Sturm–Liouville problem — an eigenvalue problem for $\frac{d}{dx}\!\big(p\,\frac{dy}{dx}\big)+q\,y=-\lambda w\,y$ with boundary conditions — became the foundation for Fourier analysis on general domains and for the mathematics of quantum mechanics a century later.

🧠 Intuition

Think of $y''$ as acceleration and $y$ as position. The homogeneous equation $y'' + p y' + q y = 0$ is a competition between three effects: the $qy$ term is a restoring force pulling $y$ back toward zero (a spring), the $py'$ term is friction that drains energy, and $y''$ is the inertia that makes the system overshoot. When restoring dominates and friction is weak, you get oscillation; when friction dominates, you get decay without oscillation. The forcing $g(x)$ is an external push. Why exactly two constants? Because to launch a particle you must specify both where it starts and how fast it's going — position and velocity. One number isn't enough; three would be redundant. Geometrically, the solutions live in a flat (two-dimensional) plane of functions, and the initial data $(y_0, y_0')$ are coordinates picking out one point.

Figure

Damped vs. undamped solutions of

y''+py'+y=0

: the restoring term forces oscillation, the friction term

p

shrinks the envelope.

ExampleConstant-Coefficient Homogeneous Equation

Given: Solve $y'' - 5y' + 6y = 0$ with $y(0)=2$ , $y'(0)=3$ .

Step 1 — Characteristic equation. Try $y=e^{rx}$ :

12

\quad r^2 - 5r + 6 = 0 \;\Rightarrow\; (r-2)(r-3)=0 \;\Rightarrow\; r=2,\,3

Derivation hint: substitute $y=e^{rx}$ so $y'=re^{rx}$ , $y''=r^2e^{rx}$ ; factor out $e^{rx}\neq 0$ .

Step 2 — General solution. Two distinct real roots give independent exponentials:

13

\quad y(x) = c_1 e^{2x} + c_2 e^{3x}

Derivation hint: $e^{2x}$ and $e^{3x}$ have nonzero Wronskian, so they form a fundamental set per [5].

Step 3 — Apply initial conditions. $y(0)=c_1+c_2=2$ and $y'(0)=2c_1+3c_2=3$ . Subtracting twice the first from the second: $c_2 = 3 - 4 = -1$ , hence $c_1 = 3$ .

\boxed{y(x) = 3e^{2x} - e^{3x}}

(Dimensionless; $x$ in whatever unit the independent variable carries.)

ExampleDamped Driven Oscillator — Spring–Mass System

Given: A mass $m = 2\,\text{kg}$ on a spring $k = 50\,\text{N/m}$ with damping $b = 4\,\text{N}\,\cdot\,\text{s/m}$ is driven by $F(t)=10\cos(3t)\,\text{N}$ . Newton's law gives $m\ddot{x} + b\dot{x} + kx = F(t)$ , i.e.

14

\quad 2\ddot{x} + 4\dot{x} + 50x = 10\cos(3t)

Derivation hint: this is $F=ma$ with spring force $-kx$ , damping $-b\dot x$ , and drive $F(t)$ , all summed.

Step 1 — Homogeneous part. Characteristic equation $2r^2+4r+50=0 \Rightarrow r^2+2r+25=0 \Rightarrow r = -1 \pm \sqrt{1-25} = -1 \pm 4.899\,i$ . So

15

\quad x_h(t) = e^{-t}\big(c_1\cos 4.899t + c_2 \sin 4.899t\big)

Derivation hint: complex roots $\alpha\pm\beta i$ give $e^{\alpha t}(\cos\beta t,\sin\beta t)$ via Euler's formula; here $\alpha=-1,\beta=\sqrt{24}\approx4.899$ .

Step 2 — Particular (steady-state) part. Guess $x_p = A\cos 3t + B\sin 3t$ . Then $\ddot x_p = -9x_p$ , and substituting into [14]:

16

\quad (50 - 2\cdot 9)(A\cos 3t + B\sin 3t) + 12(-A\sin 3t + B\cos 3t) = 10\cos 3t

Derivation hint: $2\ddot x_p = -18(A\cos+B\sin)$ , $4\dot x_p = 12(-A\sin+B\cos)$ , $50x_p$ as is; collect $\cos 3t$ and $\sin 3t$ . Matching coefficients: $32A + 12B = 10$ and $-12A + 32B = 0 \Rightarrow B = \tfrac{12}{32}A = 0.375A$ . Then $32A + 4.5A = 10 \Rightarrow A = 0.2740$ , $B = 0.1027$ .

Step 3 — Steady-state amplitude. $|x_p| = \sqrt{A^2+B^2} = \sqrt{0.0751 + 0.01056} = \sqrt{0.0857}$ .

\boxed{x_p(t) = 0.274\cos 3t + 0.103\sin 3t,\quad \text{amplitude} \approx 0.293\ \text{m}}

The transient $x_h$ decays like $e^{-t}$ within a few seconds, leaving only the steady oscillation.

⚠️ Common mistake

The general solution does not always involve elementary functions. Only special equations (constant coefficients, Euler/Cauchy equations, a handful of named forms) have closed-form solutions in terms of $e^x$ , $\sin$ , polynomials. Equations like $y'' + xy = 0$ (the Airy equation) or $y'' + (1 - n^2/x^2)y/x$ (Bessel) require new special functions. The structure theorems [5]–[8] still hold perfectly — there are still exactly two independent solutions — but you cannot necessarily write them with a finite formula.

Linear independence is not "they look different." It is governed precisely by the Wronskian [7]. Two functions can look unrelated yet be dependent on a subinterval. For solutions of the same linear ODE, Abel's identity [8] guarantees the Wronskian is all-zero or all-nonzero, so this subtlety only bites for arbitrary function pairs.

Counterexample

"Adding two solutions always gives a solution." True for homogeneous linear equations — false in general. Consider the nonlinear equation $y'' = (y')^2$ . The function $y_1 = -\ln(x)$ satisfies it ( $y_1' = -1/x$ , $y_1'' = 1/x^2 = (y_1')^2$ ✓). So does any vertical shift $y_1 + c$ . But $2y_1 = -2\ln x$ gives $y'' = 2/x^2$ while $(y')^2 = 4/x^2$ — not equal. Superposition fails because the equation is nonlinear; scaling a solution by 2 does not yield a solution. This is exactly why the clean theory of this article applies only to linear second-order equations.

connection

Second-order equations sit at the crossroads of several theories. They are a special case of Systems of First-Order ODEs via the reduction in [9], which exposes the matrix-exponential viewpoint. The constant-coefficient case is solved by the Characteristic Equation, reducing analysis to polynomial roots. Boundary-value versions feed into Sturm–Liouville Theory, whose eigenfunction expansions generalize Fourier Series. Partial differential equations such as the Wave Equation, Heat Equation, and Laplace's Equation reduce — via separation of variables — to coupled second-order ODEs, making this topic the engine room of mathematical physics. Solution techniques connect to the Laplace Transform, which turns the differential operator into algebra, and to Power Series Solutions for variable-coefficient cases.

🔬 Where this appears

⏳ History

- 1690s — Jacob and Johann Bernoulli solve specific second-order equations arising from the catenary, brachistochrone, and elastic curves, treating them ad hoc.

- 1739–1743 — Leonhard Euler systematically solves the constant-coefficient linear equation, introducing the substitution

y=e^{rx}

and the characteristic equation [12].

- 1743 — Jean d'Alembert derives and solves the wave equation for a vibrating string, reducing it to second-order ODEs via what later became separation of variables.

- 1750s — Euler and Daniel Bernoulli debate the nature of string vibrations, seeding the theory of superposition and trigonometric series.

- 1799 — Euler's "variation of parameters" idea (anticipated by Lagrange) for inhomogeneous equations is fully developed, yielding [10]–[11].

- 1824 — Niels Henrik Abel establishes the Wronskian identity [8], linking the determinant to the first-derivative coefficient.

- 1836–1837 — Charles Sturm and Joseph Liouville publish the comparison and oscillation theorems, founding qualitative theory and eigenvalue problems.

- 1858 — Bernhard Riemann and contemporaries study the hypergeometric and Bessel equations, formalizing special-function solutions of variable-coefficient cases.

- 1882 — Gaston Floquet extends the theory to periodic coefficients (Floquet theory), crucial for stability of mechanical and quantum systems.

open-problem

Singular points and the boundary of theory. The existence–uniqueness guarantee [4] requires $p,q,g$ continuous — i.e. $a_2(x)\neq 0$ . At a singular point where $a_2=0$ (e.g. $x=0$ in $x^2y''+xy'+(x^2-\nu^2)y=0$ , the Bessel equation), solutions may blow up, fail to be unique, or branch. The Frobenius method classifies regular singular points, but irregular singular points (where the singularity is "too strong") still resist general closed-form treatment, and the asymptotic behavior of solutions near them is studied by resurgence theory and exact WKB analysis — active research areas. For nonlinear second-order equations, no clean structure theorem exists: solutions can exhibit finite-time blow-up, chaotic sensitivity (e.g. the driven, damped pendulum $\ddot\theta + b\dot\theta + \sin\theta = F\cos\omega t$ ), and the absence of a superposition principle. Determining which nonlinear second-order ODEs possess the Painlevé property (movable singularities only as poles) classifies the six Painlevé transcendents — special functions whose deeper properties remain a subject of ongoing study in integrable systems.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also