Linear Second-Order ODEs

⚡ Key Takeaways

One-line statement: A linear second-order ODE has the form $a(x)y'' + b(x)y' + c(x)y = f(x)$ , where the unknown function and its derivatives appear only to the first power and are not multiplied together. Key result in words: The general solution is the sum of the general solution of the homogeneous equation (a two-dimensional vector space spanned by two linearly independent solutions) plus any single particular solution of the full equation. Where it applies: Mechanical vibrations, electrical circuits, quantum mechanics, heat conduction, wave propagation, beam deflection, and virtually every linear physical system with inertia. Who/when: Foundations laid by Euler, d'Alembert, and Lagrange (18th century); existence theory by Cauchy, Lipschitz, and Picard (19th century).

💡 The Core Insight

Linearity converts a differential equation into linear algebra. Because the operator $L[y] = a y'' + b y' + c y$ satisfies $L[\alpha y_1 + \beta y_2] = \alpha L[y_1] + \beta L[y_2]$ , the homogeneous solution set is the kernel of a linear map and is therefore a vector space. For a second-order equation that kernel is exactly two-dimensional. Everything else — superposition, the role of the Wronskian, the general solution = homogeneous + particular structure — is a direct consequence of this single algebraic fact.

DefinitionLinear Second-Order ODE

A linear second-order ODE on an interval $I$ is any equation expressible as

1

\quad a(x)\,y'' + b(x)\,y' + c(x)\,y = f(x),

Derivation hint: This is the most general form in which $y, y', y''$ appear to the first power with no products among them; coefficients may depend on $x$ but not on $y$ . where:

$y = y(x)$ is the unknown function,
$a(x), b(x), c(x)$ are the coefficient functions (with $a(x) \neq 0$ on $I$ ),
$f(x)$ is the forcing term (or source/inhomogeneity).

The equation is homogeneous if $f(x) \equiv 0$ , and inhomogeneous (nonhomogeneous) otherwise. Dividing by $a(x)$ gives the standard form

2

\quad y'' + p(x)\,y' + q(x)\,y = g(x), \qquad p = \tfrac{b}{a},\; q = \tfrac{c}{a},\; g = \tfrac{f}{a}.

Derivation hint: Valid wherever $a(x)\neq 0$ ; points where $a$ vanishes are singular points requiring separate analysis. If the coefficients $p, q$ are constants, the equation has constant coefficients; otherwise it has variable coefficients.

DefinitionDifferential Operator and Linearity

Define the linear differential operator

3

\quad L \equiv \frac{d^2}{dx^2} + p(x)\frac{d}{dx} + q(x), \qquad L[y] = y'' + p y' + q y.

Derivation hint: $L$ acts on the vector space $C^2(I)$ ; linearity follows because differentiation is linear and multiplication by a fixed function is linear. Then $L[\alpha y_1 + \beta y_2] = \alpha L[y_1] + \beta L[y_2]$ for all scalars $\alpha,\beta$ and twice-differentiable $y_1, y_2$ .

TheoremExistence and Uniqueness (Cauchy–Picard)

If $p(x)$ , $q(x)$ , and $g(x)$ are continuous on an open interval $I$ containing $x_0$ , then for any prescribed values $y_0, y_0'$ the initial value problem

4

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

Derivation hint: Reduce to a first-order system $\mathbf{u}' = A(x)\mathbf{u} + \mathbf{b}(x)$ with $\mathbf{u}=(y,y')^\top$ ; apply the Picard–Lindelöf theorem to the Lipschitz right-hand side. has a unique solution existing on the entire interval $I$ .

TheoremStructure of the Solution Set

The homogeneous equation $L[y] = 0$ has a solution set that is a two-dimensional vector space. If $y_1, y_2$ are two linearly independent solutions (a fundamental set), then every homogeneous solution is

5

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

Derivation hint: The map $y \mapsto (y(x_0), y'(x_0))$ from the kernel of $L$ to $\mathbb{R}^2$ is a linear isomorphism by existence-uniqueness; hence the kernel has dimension 2. The general solution of the inhomogeneous equation is

6

\quad y = y_h + y_p = c_1 y_1 + c_2 y_2 + y_p,

Derivation hint: If $L[y_p]=g$ and $L[Y]=g$ , then $L[Y-y_p]=0$ , so $Y-y_p$ lies in the homogeneous space; rearrange. where $y_p$ is any particular solution of $L[y] = g$ .

DefinitionWronskian and Linear Independence

The Wronskian of two solutions is

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: The determinant tests whether the constants $c_1,c_2$ matching given $(y, y')$ data are uniquely solvable; nonzero determinant ⇒ invertible coefficient matrix. Two solutions of the same equation are linearly independent on $I$ if and only if $W \neq 0$ at some (equivalently every) point of $I$ . 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 $W$ , use the ODE to replace $y_i''$ by $-p y_i' - q y_i$ ; cancellation yields $W' = -p\,W$ , a separable first-order ODE. A profound consequence: the Wronskian is either identically zero or never zero — it cannot vanish at isolated points for solutions of the same linear equation.

ProofConstant-Coefficient Homogeneous Solution via the Characteristic Equation

Consider $y'' + p y' + q y = 0$ with constant $p, q$ .

Step 1. Seek exponential solutions $y = e^{rx}$ , motivated by the fact that $e^{rx}$ reproduces itself under differentiation. Then $y' = r e^{rx}$ , $y'' = r^2 e^{rx}$ .

Step 2. Substitute:

9

\quad (r^2 + p r + q)\,e^{rx} = 0.

Derivation hint: Factor out the never-zero $e^{rx}$ ; the bracket must vanish.

Step 3. Since $e^{rx} \neq 0$ , we require the characteristic (auxiliary) equation

10

\quad r^2 + p r + q = 0, \qquad r_{1,2} = \frac{-p \pm \sqrt{p^2 - 4q}}{2}.

Derivation hint: Quadratic formula; the discriminant $\Delta = p^2 - 4q$ governs the case structure.

Step 4 — Case analysis.

Distinct real roots ( $\Delta > 0$ ): $y = c_1 e^{r_1 x} + c_2 e^{r_2 x}$ . Independence: $W = (r_2 - r_1)e^{(r_1+r_2)x} \neq 0$ .

Repeated root ( $\Delta = 0$ , $r_1 = r_2 = r$ ): one solution $e^{rx}$ ; a second independent solution is $x e^{rx}$ , found by reduction of order:

11

\quad y_2 = x\,e^{rx}, \qquad L[x e^{rx}] = (2r + p)e^{rx} = 0 \text{ since } r = -p/2.

Derivation hint: Try $y_2 = v(x)e^{rx}$ ; substitution forces $v'' = 0$ , so $v = x$ (taking the nontrivial part).

Complex conjugate roots ( $\Delta < 0$ , $r = \alpha \pm i\beta$ ): using Euler's formula $e^{i\beta x} = \cos\beta x + i\sin\beta x$ , the real fundamental set is

12

\quad y = e^{\alpha x}\left(c_1 \cos\beta x + c_2 \sin\beta x\right).

Derivation hint: Take real and imaginary parts of the complex solution $e^{(\alpha+i\beta)x}$ ; both solve the real ODE because $L$ has real coefficients. This exhausts all cases and produces a two-dimensional solution space, confirming the Structure Theorem. $\blacksquare$

∎

ProofVariation of Parameters — Finding a Particular Solution

Given a fundamental set $\{y_1, y_2\}$ of $L[y]=0$ , seek $y_p = u_1(x)y_1 + u_2(x)y_2$ .

Step 1. Impose the convenient constraint $u_1' y_1 + u_2' y_2 = 0$ to keep the algebra clean. Then $y_p' = u_1 y_1' + u_2 y_2'$ .

Step 2. Differentiate again and substitute into $L[y_p] = g$ . The homogeneous terms cancel, leaving

13

\quad u_1' y_1' + u_2' y_2' = g.

Derivation hint: $L[y_1]=L[y_2]=0$ eliminates all undifferentiated $u_i$ terms; only the derivatives of $u_i$ survive.

Step 3. Solve the linear system (constraint + [13]) by Cramer's rule using the Wronskian:

14

\quad u_1' = -\frac{y_2\, g}{W}, \qquad u_2' = \frac{y_1\, g}{W}.

Derivation hint: The $2\times2$ coefficient matrix is exactly the Wronskian matrix; invert it.

Step 4. Integrate to obtain the particular solution:

15

\quad y_p(x) = -y_1\!\int\!\frac{y_2\, g}{W}\,dx + y_2\!\int\!\frac{y_1\, g}{W}\,dx.

Derivation hint: Antidifferentiate [14]; constants of integration only add homogeneous solutions, already absorbed into $y_h$ . This always works whenever $\{y_1,y_2\}$ is known, regardless of the form of $g$ . $\blacksquare$

∎

LE

Leonhard Euler

1707–1783

Euler was the architect of the exponential-substitution method for linear ODEs with constant coefficients. In his Institutiones calculi integralis (1768–1770) he systematized the use of $y = e^{rx}$ and explicitly handled repeated and complex roots, employing the formula $e^{i\theta} = \cos\theta + i\sin\theta$ that bears his name to convert complex exponentials into real trigonometric solutions. He also introduced the technique now called the integrating factor and studied the equidimensional (Euler–Cauchy) equation $x^2 y'' + a x y' + b y = 0$ . His correspondence with d'Alembert and Daniel Bernoulli on the vibrating string drove much of the theory forward, and his notation for differentials and functions remains standard.

🧠 Intuition

Think of a linear second-order ODE as describing a mass on a spring with possible friction and an external push. The term $y''$ is acceleration (inertia), $p y'$ is velocity-proportional damping, $q y$ is a restoring force, and $g$ is the external driving force. The characteristic roots are the system's "DNA": their real parts tell you whether disturbances grow or decay; their imaginary parts tell you whether the system oscillates.

Two negative real roots → sluggish, overdamped return to rest (think a door closer with stiff hydraulics).
A repeated negative root → critically damped, the fastest non-oscillating return.
Complex roots with negative real part → underdamped ringing that fades.
Pure imaginary roots → undamped perpetual oscillation.

The two-dimensional solution space corresponds to the two pieces of information a Newtonian system needs to predict the future: initial position and initial velocity. Once you fix those, the trajectory is locked in forever — that is uniqueness made physical.

Figure

Damping regimes of

y'' + 2\zeta\omega_0 y' + \omega_0^2 y = 0

for fixed initial displacement

ExampleUnderdamped Mechanical Oscillator

Given: A mass–spring–damper with $m = 2\ \text{kg}$ , damping $c = 4\ \text{N}\,\cdot\,\text{s/m}$ , stiffness $k = 50\ \text{N/m}$ , released from rest at displacement $x(0) = 0.1\ \text{m}$ , $x'(0) = 0$ .

Equation: $m x'' + c x' + k x = 0 \Rightarrow 2x'' + 4x' + 50x = 0 \Rightarrow x'' + 2x' + 25x = 0$ .

Step 1 — Characteristic equation: $r^2 + 2r + 25 = 0$ . Discriminant $\Delta = 4 - 100 = -96 < 0$ .

Step 2 — Roots: $r = \dfrac{-2 \pm \sqrt{-96}}{2} = -1 \pm \sqrt{24}\,i = -1 \pm 2\sqrt{6}\,i$ . So $\alpha = -1\ \text{s}^{-1}$ , $\beta = 2\sqrt{6} \approx 4.899\ \text{rad/s}$ .

Step 3 — General solution: $x(t) = e^{-t}\big(c_1 \cos\beta t + c_2 \sin\beta t\big)$ .

Step 4 — Apply initial conditions: $x(0) = c_1 = 0.1$ . $x'(t) = e^{-t}\big[(-c_1 + \beta c_2)\cos\beta t + (-c_2 - \beta c_1)\sin\beta t\big]$ ; setting $x'(0) = 0$ : $-c_1 + \beta c_2 = 0 \Rightarrow c_2 = c_1/\beta = 0.1/4.899 = 0.0204$ .

Answer:

\boxed{x(t) = e^{-t}\big(0.100\cos(4.899\,t) + 0.0204\sin(4.899\,t)\big)\ \text{m}}

The amplitude decays as $e^{-t}$ (time constant $1\ \text{s}$ ) while ringing at $\approx 0.78\ \text{Hz}$ .

ExampleForced RLC Circuit — Variation of Parameters

Given: Solve $y'' + 4y = 8\sin(t)$ with $y(0) = 0$ , $y'(0) = 0$ (a circuit driven off-resonance; natural frequency $2$ , drive frequency $1$ ).

Step 1 — Homogeneous solution: $r^2 + 4 = 0 \Rightarrow r = \pm 2i$ , so $y_h = c_1\cos 2t + c_2 \sin 2t$ . Here $y_1 = \cos 2t,\ y_2 = \sin 2t$ , $W = y_1 y_2' - y_2 y_1' = 2\cos^2 2t + 2\sin^2 2t = 2$ .

Step 2 — Particular solution by method of undetermined coefficients (faster here): try $y_p = A\sin t + B\cos t$ . Then $y_p'' = -A\sin t - B\cos t$ , so $y_p'' + 4y_p = 3A\sin t + 3B\cos t = 8\sin t$ . Match: $3A = 8 \Rightarrow A = 8/3$ , $B = 0$ .

Step 3 — General solution: $y = c_1\cos 2t + c_2\sin 2t + \tfrac{8}{3}\sin t$ .

Step 4 — Initial conditions: $y(0) = c_1 = 0$ . $y'(t) = -2c_1\sin 2t + 2c_2\cos 2t + \tfrac{8}{3}\cos t$ ; $y'(0) = 2c_2 + \tfrac{8}{3} = 0 \Rightarrow c_2 = -\tfrac{4}{3}$ .

Answer:

\boxed{y(t) = -\tfrac{4}{3}\sin 2t + \tfrac{8}{3}\sin t}

The motion is a superposition of the natural mode (frequency 2) and the steady driven response (frequency 1) — a clean beating pattern with no growth because the drive is off resonance.

⚠️ Common mistake

The sum of two solutions of an inhomogeneous equation is generally NOT a solution. Superposition holds only for the homogeneous equation. If $L[y_1] = g$ and $L[y_2] = g$ , then $L[y_1 + y_2] = 2g \neq g$ . Superposition of forcing terms does work in this sense: if $L[y_a] = g_a$ and $L[y_b] = g_b$ , then $L[y_a + y_b] = g_a + g_b$ . Keep these two statements distinct.

"Linear" does not mean "solvable in closed form." Linearity guarantees structural theorems (two-dimensional solution space, superposition), but variable-coefficient equations like Bessel's or Legendre's require special functions or series — there is no elementary formula in general.

Counterexample

Two solutions that vanish together but are still independent? No — and here is why the Wronskian matters. Consider the functions $f(x) = x^2$ and $g(x) = x|x|$ on $(-1,1)$ . Their Wronskian is identically zero, yet they are linearly independent as abstract functions. This does not contradict the theory because they are not both solutions of the same linear second-order ODE with continuous coefficients on that interval — no such equation exists for which $x|x|$ is a $C^2$ solution at the origin. The Wronskian's "all-or-nothing" property (Abel's identity) holds only for genuine solutions of one linear equation; it is precisely what distinguishes solution functions from arbitrary functions.

connection

Linear second-order ODEs sit at a crossroads of analysis and algebra. They are the simplest nontrivial case of Linear Differential Equations of arbitrary order, and reduce to a planar Systems of First-Order ODEs via $\mathbf{u} = (y, y')^\top$ , linking them to matrix exponentials and eigenvalue theory. The constant-coefficient case is the natural home of the Laplace Transform, which algebraizes the equation into a rational function of $s$ . Variable-coefficient cases with singular points open into the Frobenius Method and the theory of Special Functions (Bessel, Legendre, Hermite). The eigenvalue problem $L[y] = \lambda y$ with boundary conditions is the gateway to Sturm–Liouville Theory and the spectral decomposition underlying Fourier Series.

🔬 Where this appears

⏳ History

- 1690s — The Bernoulli family and Leibniz solve early special second-order equations arising from the catenary and elastica problems.

- 1739 — Euler develops the exponential-substitution method for linear constant-coefficient equations and treats repeated and complex roots systematically.

- 1743 — Euler publishes the characteristic-equation technique; the term "characteristic" enters the lexicon. d'Alembert's work on the vibrating string (1747) supplies key physical motivation.

- 1760s–1770s — Lagrange invents variation of parameters and adjoint methods, generalizing particular-solution construction to arbitrary forcing.

- 1827 — Niels Henrik Abel proves Abel's identity for the Wronskian, clarifying the all-or-nothing nature of linear independence among solutions.

- 1836–1837 — Charles Sturm and Joseph Liouville develop the spectral theory of

(p y')' + (q + \lambda w)y = 0

, founding Sturm–Liouville theory and oscillation theorems.

- 1838 — Friedrich Bessel's functions and the Bessel equation become central to variable-coefficient problems in physics and astronomy.

- 1868 — The Frobenius series method (refined later by Frobenius, 1873) provides solutions near regular singular points.

- 1890s — Picard and Lindelöf establish the modern existence-uniqueness theory via successive approximations, placing the entire subject on rigorous footing.

open-problem

While the linear theory is essentially complete, several deep issues persist at its boundaries. Closed-form solvability is governed by differential Galois theory (Picard–Vessiot), which determines exactly when a linear ODE has solutions expressible in elementary or Liouvillian terms — but deciding this algorithmically for high-order or parametric families remains intricate. Singular points where $a(x) = 0$ (irregular singularities especially) generate solutions with essential singularities and Stokes phenomena, whose asymptotic behavior connects to resurgence theory and is still actively studied. The transition from linear to weakly nonlinear systems (e.g., the Duffing or van der Pol equations) shows that linear superposition is a fragile idealization; perturbation series can diverge, and chaos appears. Finally, the spectral theory of singular Sturm–Liouville operators on unbounded domains — central to quantum scattering — raises subtle questions about continuous spectra, self-adjoint extensions, and limit-point/limit-circle classification that blend ODE theory with functional analysis.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also