Ordinary Differential Equations

⚡ Key Takeaways

Ordinary Differential Equations (ODEs) are equations relating a function of a single variable to its derivatives. One-line statement: An ODE asks "what function has this relationship between itself and its rates of change?" Key result: Under mild conditions (continuity + Lipschitz), a unique solution exists locally for any initial condition (Picard–Lindelöf). Where it applies: physics, engineering, biology, economics — anywhere a system evolves deterministically in one variable (usually time). Who/when: born with Newton and Leibniz (1670s–1680s); systematized by Euler, the Bernoullis, Cauchy, and Lipschitz through the 18th–19th centuries.

💡 The Core Insight

A differential equation does not hand you a function — it hands you a rule for the slope at every point. The solution is the unique curve that everywhere obeys that rule and threads through a chosen starting point. Geometrically, a first-order ODE $y' = f(x,y)$ paints a direction field on the plane: a tiny arrow of slope $f(x,y)$ at each point. Solving the ODE means finding curves tangent to every arrow they pass through. Existence is the claim that such curves always exist; uniqueness is the claim that through each point passes exactly one.

DefinitionOrdinary Differential Equation

An ordinary differential equation of order $n$ in the unknown function $y(x)$ is a relation

1

\quad F\!\left(x,\, y,\, \frac{dy}{dx},\, \frac{d^2y}{dx^2},\, \dots,\, \frac{d^n y}{dx^n}\right) = 0,

where $x$ is the independent variable, $y = y(x)$ is the dependent variable (the unknown), $y^{(k)} = d^k y/dx^k$ is the $k$ -th derivative, and $F$ is a given function. The order $n$ is the highest derivative appearing. The equation is explicit (or in normal form) if it can be solved for the top derivative:

2

\quad y^{(n)} = f\!\left(x, y, y', \dots, y^{(n-1)}\right).

Derivation hint: Equation [2] is just [1] solved algebraically for $y^{(n)}$ where the implicit function theorem permits; not every implicit ODE admits a global normal form. A solution on an interval $I \subseteq \mathbb{R}$ is an $n$ -times differentiable function $\varphi: I \to \mathbb{R}$ that turns [1] into an identity for all $x \in I$ .

DefinitionLinear ODE

An ODE is linear if it can be written

3

\quad a_n(x)\,y^{(n)} + a_{n-1}(x)\,y^{(n-1)} + \cdots + a_1(x)\,y' + a_0(x)\,y = g(x),

with the unknown $y$ and its derivatives appearing only to the first power and not multiplied together. The coefficients $a_k(x)$ depend on $x$ alone. If $g(x) \equiv 0$ the equation is homogeneous; otherwise inhomogeneous (or forced). If all $a_k$ are constants it has constant coefficients. Derivation hint: Linearity means the left side defines a linear operator $L[y] = \sum_k a_k(x)\, y^{(k)}$ , so $L[\alpha y_1 + \beta y_2] = \alpha L[y_1] + \beta L[y_2]$ .

TheoremPicard–Lindelöf Existence and Uniqueness

Consider the first-order initial value problem (IVP)

4

\quad y' = f(x,y), \qquad y(x_0) = y_0.

Derivation hint: This is [2] for $n=1$ together with one initial datum that pins down the integration constant. If $f$ is continuous on a rectangle $R = \{|x - x_0| \le a,\ |y - y_0| \le b\}$ and Lipschitz continuous in $y$ there — meaning there is a constant $L$ with

5

\quad |f(x, y_1) - f(x, y_2)| \le L\,|y_1 - y_2| \quad \text{for all } (x,y_1),(x,y_2)\in R,

then there exists $h > 0$ such that the IVP [4] has a unique solution $y(x)$ on the interval $|x - x_0| \le h$ , where $h = \min\!\left(a, \, b/M\right)$ and $M = \max_R |f|$ . Derivation hint: $M$ bounds the slope, so over a horizontal span $h$ the solution can rise at most $Mh \le b$ , staying inside $R$ ; this is why $h \le b/M$ .

ProofPicard Iteration — sketch of existence and uniqueness

Step 1 — Reformulate as an integral equation. Integrating [4] from $x_0$ to $x$ converts the differential problem into a fixed-point problem:

6

\quad y(x) = y_0 + \int_{x_0}^{x} f\!\left(t, y(t)\right)\,dt.

Derivation hint: Apply the Fundamental Theorem of Calculus to $y'=f$ ; the integral equation encodes both the ODE and the initial condition in one stroke.

Step 2 — Build the Picard iterates. Define a sequence of approximations starting from the constant $y_0$ :

7

\quad y_{k+1}(x) = y_0 + \int_{x_0}^{x} f\!\left(t, y_k(t)\right)\,dt, \qquad y_0(x) = y_0.

Derivation hint: This is the Banach fixed-point iteration $y_{k+1} = T[y_k]$ applied to the operator $T$ defined by the right side of [6].

Step 3 — Show the operator is a contraction. Using the Lipschitz bound [5], the distance between successive iterates shrinks:

8

\quad \|y_{k+1} - y_k\|_\infty \le \frac{(Lh)^{k}}{k!}\,\|y_1 - y_0\|_\infty .

Derivation hint: Iterate the inequality $|y_{k+1}-y_k| \le L\int_{x_0}^x |y_k - y_{k-1}|\,dt$ ; each integration introduces another factor of $|x-x_0| \le h$ and a falling factorial, giving the exponential-series bound.

Step 4 — Conclude convergence. Because $\sum_k (Lh)^k/k! = e^{Lh} < \infty$ , the series $\sum (y_{k+1}-y_k)$ converges absolutely and uniformly, so $y_k \to y$ uniformly. The limit satisfies [6], hence solves [4].

Step 5 — Uniqueness. Suppose $y$ and $z$ both solve [6]. Then $|y-z|(x) \le L\int_{x_0}^x |y-z|\,dt$ . By Grönwall's inequality this forces $|y-z| \equiv 0$ . $\blacksquare$

∎

⚠️ Continuity alone gives existence but NOT uniqueness

Peano's theorem guarantees a solution exists whenever $f$ is merely continuous. But continuity is too weak for uniqueness — the Lipschitz condition [5] is essential. See the counterexample below.

AC

Augustin-Louis Cauchy

1789–1857

Cauchy transformed differential equations from a collection of solving tricks into a rigorous theory. Before him, mathematicians asked "what is the formula for the solution?"; Cauchy asked the deeper question "does a solution exist, and is it unique?" — the Cauchy problem (an ODE plus initial data). His method of majorants (1820s) gave the first general existence proof for analytic $f$ . Rudolf Lipschitz (1876) and Émile Picard (1890) later relaxed and streamlined the hypotheses into the iteration scheme above. Cauchy's insistence on $\varepsilon$ – $\delta$ rigor reshaped all of analysis, but his work on ODEs is where the modern notion of a well-posed problem — existence, uniqueness, continuous dependence — was effectively born.

🧠 Intuition

Think of $y' = f(x,y)$ as a current in a river. At every point of the $xy$ -plane the function $f$ specifies a flow direction (a slope). Drop a leaf at $(x_0, y_0)$ : it traces out the solution curve, always moving tangent to the local flow. The direction field (or slope field) is a snapshot of all these little arrows. A solution is a streamline. The Lipschitz condition says the flow doesn't change too violently from point to point — if it did, two streamlines could merge or split (non-uniqueness), like water at a singular drain. For higher-order equations the picture lifts into phase space: an $n$ -th order ODE becomes a first-order system in $n$ variables $(y, y', \dots, y^{(n-1)})$ , and a single solution becomes a trajectory through that higher-dimensional space, where again exactly one trajectory passes through each point.

Figure

Direction field of

y' = y

(exponential growth) with three solution curves through different starting points

ExampleRadioactive decay — a first-order linear ODE

Given: A sample of carbon-14 decays at a rate proportional to the amount present, with half-life $T_{1/2} = 5730$ years. Initial mass $N_0 = 10.0$ g. Find the mass after $2000$ years.

Step 1 — Model. "Rate proportional to amount" means

9

\quad \frac{dN}{dt} = -k\,N, \qquad N(0) = N_0.

Derivation hint: The minus sign encodes decay; $k>0$ is the decay constant with units $\text{time}^{-1}$ .

Step 2 — Separate variables and integrate.

10

\quad \int \frac{dN}{N} = -k\int dt \;\Longrightarrow\; \ln N = -kt + C \;\Longrightarrow\; N(t) = N_0\,e^{-kt}.

Derivation hint: Separable form $dN/N = -k\,dt$ ; exponentiate and use $N(0)=N_0$ to fix $C$ .

Step 3 — Find $k$ from the half-life. $N(T_{1/2}) = N_0/2$ gives

11

\quad k = \frac{\ln 2}{T_{1/2}} = \frac{0.6931}{5730} = 1.2097\times 10^{-4}\ \text{yr}^{-1}.

Derivation hint: Set $e^{-kT_{1/2}} = \tfrac12$ and solve.

Step 4 — Evaluate at $t = 2000$ yr. $kt = 1.2097\times10^{-4}\times 2000 = 0.24194$ , so $e^{-0.24194} = 0.78511$ .

\boxed{N(2000) = 10.0 \times 0.78511 = 7.85\ \text{g}}

ExampleDamped mass–spring — a second-order linear constant-coefficient ODE

Given: Mass $m = 2.0$ kg, damping $c = 4.0$ N·s/m, spring constant $k = 10.0$ N/m. Released from $x(0) = 0.10$ m with $x'(0) = 0$ . Find the motion.

Step 1 — Newton's law.

12

\quad m\,x'' + c\,x' + k\,x = 0 \;\Longrightarrow\; 2x'' + 4x' + 10x = 0.

Derivation hint: Sum of inertial, damping, and restoring forces equals zero (free vibration).

Step 2 — Characteristic equation. Substitute $x = e^{rt}$ :

13

\quad 2r^2 + 4r + 10 = 0 \;\Longrightarrow\; r = \frac{-4 \pm \sqrt{16 - 80}}{4} = -1 \pm 2i.

Derivation hint: $e^{rt}$ turns derivatives into powers of $r$ ; the discriminant $<0$ signals oscillation (underdamping).

Step 3 — General solution (complex roots $\alpha \pm \beta i$ give decaying oscillation).

14

\quad x(t) = e^{-t}\big(A\cos 2t + B\sin 2t\big).

Derivation hint: Real and imaginary parts of $e^{(\alpha+\beta i)t}$ form a real basis $e^{\alpha t}\cos\beta t,\ e^{\alpha t}\sin\beta t$ .

Step 4 — Apply initial conditions. $x(0)=A=0.10$ . Differentiate and set $t=0$ : $x'(0) = -A + 2B = 0 \Rightarrow B = A/2 = 0.05$ .

\boxed{x(t) = e^{-t}\big(0.10\cos 2t + 0.05\sin 2t\big)\ \text{m}}

The motion is a decaying oscillation with angular frequency $2$ rad/s and decay rate $1\ \text{s}^{-1}$ .

⚠️ Five frequent errors

1. "More initial conditions are always better." An $n$ -th order ODE needs exactly $n$ conditions. Too few leaves free constants; too many is generally over-determined (no solution). 2. "The general solution + one condition = unique answer." Only if those conditions are initial (all at one point). Boundary conditions (at two points) can yield zero, one, or infinitely many solutions. 3. "Separation of variables works for any first-order ODE." It works only when $f(x,y)$ factors as $g(x)h(y)$ . 4. "Adding particular solutions of different forcings is illegitimate." For linear equations superposition holds; for nonlinear ones it categorically fails. 5. "A solution exists for all time." Even nice ODEs can blow up in finite time (see below). Picard–Lindelöf is only local.

CounterexampleContinuity without Lipschitz: non-uniqueness

Consider

15

\quad y' = 3\,y^{2/3}, \qquad y(0) = 0.

Derivation hint: $f(y)=3y^{2/3}$ is continuous everywhere but its $y$ -derivative $2y^{-1/3}$ blows up at $y=0$ , so it fails the Lipschitz condition [5] there. This IVP has infinitely many solutions: $y \equiv 0$ , and for any $c \ge 0$ ,

y(x) = \begin{cases} 0, & x \le c,\\ (x-c)^3, & x > c.\end{cases}

Each solution "rests" at zero for an arbitrary time $c$ then takes off. Uniqueness fails precisely because Lipschitz fails.

CounterexampleFinite-time blow-up

The harmless-looking IVP

16

\quad y' = y^2, \qquad y(0) = 1

Derivation hint: Separable: $\int dy/y^2 = \int dx \Rightarrow -1/y = x + C$ , with $C=-1$ from the initial datum. has solution $y(x) = 1/(1-x)$ , which escapes to $+\infty$ as $x \to 1^-$ . The solution simply ceases to exist beyond $x=1$ , even though $f(x,y)=y^2$ is smooth everywhere — confirming that Picard–Lindelöf only promises a local interval.

connection

ODEs sit at the crossroads of nearly all of analysis and applied mathematics.

Linear Algebra — solving a linear system of ODEs $\mathbf{y}' = A\mathbf{y}$ reduces to finding eigenvalues and eigenvectors of $A$ ; the solution is $\mathbf{y}=e^{At}\mathbf{y}_0$ via the matrix exponential.
<a class="wiki-link" href="/wiki/laplace-transform">Laplace Transform</a> — converts linear constant-coefficient ODEs (and their initial conditions) into algebraic equations, then inverts.
Power Series Methods — solve ODEs with variable coefficients (e.g. Bessel's, Legendre's equation) by expanding the solution as a series.
<a class="wiki-link" href="/wiki/partial-differential-equations">Partial Differential Equations</a> — separation of variables reduces a PDE to several coupled ODEs; ODE theory is the foundation.
Dynamical Systems — the qualitative, long-term study of ODE solutions (fixed points, stability, chaos) without solving them explicitly.
Numerical Analysis — when no closed form exists, methods like Euler and Runge–Kutta approximate solutions step by step.

🔬 Where this appears

⏳ History

- 1671–1676 — Isaac Newton, in Method of Fluxions, solves the first differential equations using infinite series and his "fluxional" notation; keeps the work largely private.

- 1684–1693 — Gottfried Leibniz publishes the

dy/dx

notation and the term aequatio differentialis; with the Bernoullis he develops separation of variables and the method for first-order linear equations.

- 1696 — Johann Bernoulli poses the brachistochrone problem, sparking the calculus of variations and the systematic study of ODEs across Europe.

- 1739–1755 — Leonhard Euler systematizes linear ODEs with constant coefficients via the characteristic equation, introduces the integrating factor, and pioneers numerical (Euler) methods.

- 1820s — Augustin-Louis Cauchy proves the first general existence theorem (method of majorants) and frames the Cauchy problem, demanding rigor over formula-hunting.

- 1838 — Joseph Liouville and others establish that most ODEs (e.g. the Riccati equation generically) have no solution in elementary functions — the analogue of unsolvable algebraic equations.

- 1876 — Rudolf Lipschitz isolates the condition [5] guaranteeing uniqueness.

- 1890 — Émile Picard perfects the method of successive approximations [7], giving the clean Picard–Lindelöf theorem still taught today.

- 1890s — Henri Poincaré founds the qualitative theory of ODEs and dynamical systems, studying stability and the three-body problem, anticipating chaos theory.

open-problem

Several deep difficulties persist beyond the undergraduate core. (1) Closed-form solvability: there is no algorithm to decide whether a general ODE has an elementary-function solution; differential Galois theory (Picard–Vessiot) partly addresses the linear case but the nonlinear landscape is wide open. (2) Global existence vs. blow-up: for nonlinear systems (e.g. those arising from fluid dynamics), determining whether solutions remain smooth for all time or develop singularities is a major open question — the Navier–Stokes regularity problem (a PDE, but rooted in ODE-style energy estimates) is a Clay Millennium Prize problem. (3) Sensitive dependence and chaos: even simple smooth ODEs (Lorenz system, three-body problem) exhibit dynamics so sensitive to initial conditions that long-term prediction is fundamentally impossible despite deterministic equations. (4) Stiff and singular equations: equations with widely separated time scales defy naive numerical integration, and the theory of singular perturbations remains an active research field.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also