Ordinary Differential Equations

An ordinary differential equation (ODE) is an equation relating an unknown function of a single variable to its derivatives.

10 frameworks, from 1670 to today · Analysis · Mathematics

still a working view ended · each leads to its entry below

Overview

An ordinary differential equation (ODE) is an equation relating an unknown function of a single variable to its derivatives1. If $y(t)$ is the unknown—a position, a temperature, a population, a current—the equation constrains how fast $y$ changes in terms of $y$ itself and the variable $t$. The subject of ODEs is the study of such equations: when they have solutions, what those solutions look like, how to compute them, and what they tell us about the systems they model. It is one of the oldest and most useful branches of analysis, because almost every quantitative science reaches a point where it asks how a quantity changes and wants to know where it goes.

Solving by quadrature: the seventeenth and eighteenth centuries

The subject began as a branch of calculus2. Newton and Leibniz, in the 1670s and 1680s, understood that finding a function from its rate of change is the inverse of differentiation, and the earliest ODEs were solved by recognizing them as derivatives of known quantities. The governing ambition of the next century was solvability by quadrature: expressing the solution of an ODE explicitly in terms of elementary functions and integrals—"quadratures" being integrals, from the geometric problem of finding areas.

The key early discovery was the separation of variables: an equation of the form $dy/dx = f(x)g(y)$ can be rearranged so that all $y$'s sit on one side and all $x$'s on the other, then integrated. Leibniz and the Bernoulli brothers worked this out in the 1690s. A second landmark was the integrating factor, a multiplier that turns a certain first-order equation into an exact derivative; Leibniz found the method around 1694. The Bernoulli family—Jacob, Johann, and later Daniel—dominated the field for two generations, and Johann's student Leonhard Euler turned it into a systematic discipline.

Euler's great contribution was the linear equation with constant coefficients. For an equation like $y'' + py' + qy = 0$ with constant $p$ and $q$, Euler (building on work of d'Alembert and Daniel Bernoulli on vibrating strings) found that solutions are built from exponentials $e^{rx}$, where $r$ solves the algebraic equation $r^2 + pr + q = 0$. This reduced a problem in calculus to a problem in algebra, and it remains the first technique every student learns. The method of variation of parameters, for forcing terms, and the use of power series to solve equations with variable coefficients (Euler, and later Frobenius) extended the reach further.

By the end of the eighteenth century the field had a large catalogue of solvable types, and the question naturally arose: which equations can be solved this way, and which cannot? The answer, for second-order equations, came from elliptic functions. In the 1820s and 1830s, Abel and Jacobi studied the pendulum equation and found that its solution requires elliptic integrals, which are not elementary. This was a genuine limit on quadrature, and it pushed the subject toward a different question: not what is the formula, but what can be said about the solution without one.

Existence and uniqueness: Cauchy, Lipschitz, and the meaning of a solution

The nineteenth century's central achievement was to put the subject on a rigorous footing. Augustin-Louis Cauchy, in the 1820s, introduced the idea of an initial value problem: given $y' = f(t,y)$ and a starting value $y(t0) = y0$, is there a solution, and is it unique? Cauchy proved local existence and uniqueness under a continuity condition on $f$. Rudolf Lipschitz, in 1876, sharpened the condition to what is now called the Lipschitz condition: $f$ must not change too fast in $y$, meaning there is a constant $L$ with $|f(t,y1) - f(t,y2)| \le L|y1 - y2|$. Under this condition the solution exists and is unique on some interval around $t_0$.

The proof method matters as much as the result. The Picard iteration (Émile Picard, 1890) constructs the solution as a limit of successive approximations: start with a guess, integrate the equation to get a better guess, repeat. This is a fixed-point argument, and it became the model for a great deal of later analysis. The theorem also clarified what "solution" means: a differentiable function satisfying the equation, not a formula. The Lipschitz condition is sharp in the sense that without it uniqueness can fail—the equation $y' = \sqrt{|y|}$ with $y(0)=0$ has both the zero solution and $y = t^2/4$ for $t \ge 0$1.

Linear theory and Sturm–Liouville spectral theory

Linear equations—those in which the unknown and its derivatives appear only to the first power—received a theory of their own13. The set of solutions of a homogeneous linear ODE of order $n$ forms an $n$-dimensional vector space, so every solution is a linear combination of $n$ independent ones. This superposition principle was understood by the early nineteenth century and made linear equations tractable in a way nonlinear ones are not.

The deepest development was Sturm–Liouville theory, named for Jacques Charles François Sturm and Joseph Liouville, who in the 1830s studied equations of the form $-(p(x)y')' + q(x)y = \lambda w(x)y$ on an interval with boundary conditions. They proved that the eigenvalues $\lambda$ are real and discrete, that eigenfunctions belonging to different eigenvalues are orthogonal, and that any reasonable function can be expanded in a series of eigenfunctions. This is the continuous analogue of diagonalizing a symmetric matrix, and it is the origin of much of what later became functional analysis. The vibrating string, the heat equation in a rod, and the quantum particle in a box all reduce to Sturm–Liouville problems, and the eigenfunction expansion is the tool that solves them.

Qualitative theory: Poincaré and the geometric turn

The decisive shift came from Henri Poincaré in the 1880s. Faced with the three-body problem, which resisted explicit solution, Poincaré asked a different question: what does the family of solutions look like in the plane of $(y, y')$—the phase plane? He introduced the phase portrait, a picture of the trajectories, and classified the behavior near equilibrium points (where $f = 0$): nodes, saddles, foci, centers. He showed that the qualitative picture is often stable under small changes in the equation, and he proved that some systems have no closed-form solution yet have a perfectly describable geometry.

This qualitative theory of differential equations was a change in what the subject was for4. Instead of asking for a formula, one asks for the shape of the flow: does a solution return to its start (a periodic orbit), approach an equilibrium, or wander? Poincaré's work on periodic orbits and his introduction of what became Poincaré sections—snapshots of a trajectory each time it crosses a surface—gave the field its central geometric vocabulary. His contemporary Lyapunov, in Russia, developed a parallel theory of stability, and the two traditions merged in the twentieth century.

Stability theory: Lyapunov and the question of robustness

Stability asks whether a solution that starts near an equilibrium stays near it5. An equilibrium is stable if small perturbations produce small deviations, and asymptotically stable if the deviations also die out5. For linear systems with constant coefficients, stability is decided by the eigenvalues: if all have negative real parts, the equilibrium is asymptotically stable.

Aleksandr Lyapunov, in his 1892 doctoral thesis, found a way to decide stability without solving the equation. His direct method (or second method) looks for a scalar function $V(y)$, now called a Lyapunov function, that decreases along trajectories—like a bowl whose bottom is the equilibrium65. If such a function exists, the equilibrium is stable; if it decreases strictly, asymptotically stable. The idea is a generalization of energy: a damped pendulum loses energy, so it settles. Lyapunov's method is the main tool for nonlinear stability and remains central in control theory, where one designs a controller by finding a $V$ that certifies the closed loop is stable.

Numerical integration: Euler, Runge–Kutta, and the computer

Most ODEs cannot be solved in closed form, so from the beginning practitioners computed approximate solutions. Euler's method (published in his 1768 Institutiones calculi integralis) steps forward along the tangent line: $y{n+1} = yn + h f(tn, yn)$. It is simple and, for small step $h$, converges—but slowly, with error proportional to $h$.

The great improvement came from Carl Runge (1895) and Wilhelm Kutta (1901), who built Runge–Kutta methods: each step evaluates $f$ at several intermediate points and combines them to match the Taylor expansion of the true solution to higher order78. The classical fourth-order method, with four evaluations per step and error proportional to $h^4$, became the workhorse of scientific computing. John Butcher, from the 1960s, gave the general theory of these methods through Butcher tableaux, and the order conditions that say when a method achieves a given accuracy.

The computer transformed the field. With machines, one could integrate systems of hundreds of equations—weather models, orbits, chemical kinetics—and the questions shifted to efficiency, error control, and the design of adaptive step sizes that shrink when the solution changes fast. The predictor–corrector methods (Adams–Bashforth, Adams–Moulton) and the multistep methods of the early twentieth century were re-examined and optimized for machine computation.

Stiff equations and implicit methods

A practical problem emerged in the 1950s and 1960s: many systems contain processes on wildly different time scales—a fast chemical reaction alongside a slow one, a stiff spring alongside a soft one. Such equations are called stiff79. Explicit methods like classical Runge–Kutta are forced to take tiny steps to remain stable, even when the slow behavior is smooth, making them uselessly slow98.

The remedy is implicit methods, which determine the next value by solving an equation involving it, rather than stepping forward explicitly98. The backward Euler method, $y{n+1} = yn + h f(t{n+1}, y{n+1})$, is stable for stiff problems but only first-order accurate. The backward differentiation formulas (BDF), developed by C. William Gear in 1971, and the implicit Runge–Kutta methods of the 1960s and 1970s gave higher-order stiff solvers. The price is that each step requires solving a nonlinear system, usually by Newton's method, which is why stiff solvers are more expensive per step but can take vastly larger steps. The distinction between stiff and non-stiff solvers, and the design of software that switches between them, is a defining feature of modern numerical ODE practice.

Functional analysis and the operator viewpoint

In the twentieth century, ODE theory was absorbed into functional analysis. The initial value problem $y' = Ay$, $y(0) = y0$, where $A$ is a linear operator, has the solution $y(t) = e^{tA} y0$—the matrix exponential when $A$ is a matrix, and a semigroup of operators in infinite dimensions. The Hille–Yosida theorem (1948) characterizes which operators generate such semigroups, and this framework handles not only ODEs but also partial differential equations treated as ODEs in a function space. The variation of constants formula, the resolvent, and the spectral theory of operators all became part of the ODE toolkit. This abstraction made the subject a chapter of operator theory, and it is the language in which modern linear ODE theory is written.

Dynamical systems and chaos

The qualitative tradition of Poincaré grew into the modern theory of dynamical systems. George David Birkhoff in the early twentieth century proved the Poincaré–Birkhoff theorem on periodic orbits and developed the ergodic theory of flows10. The decisive shock came from Edward Lorenz in 1963, who studied a three-equation model of convection and found that its solutions are exquisitely sensitive to initial conditions: two nearby starting points diverge exponentially. This is chaos, and it showed that even a simple deterministic ODE system can be unpredictable in practice11. The Lorenz attractor—a fractal set on which trajectories wander—became the emblem of the field.

The mathematical machinery for chaos was built in the 1960s and 1970s: hyperbolic dynamics (Smale's horseshoe, Anosov systems), bifurcation theory (how the qualitative picture changes as a parameter varies), and the Kolmogorov–Arnold–Moser (KAM) theorem on the persistence of quasi-periodic motion. The Hopf bifurcation, in which an equilibrium gives birth to a periodic orbit, and the period-doubling route to chaos (Feigenbaum, 1978) became standard. This work belongs as much to dynamical systems as to ODEs, but the ODE is its native setting: a finite system of first-order equations is exactly a flow on a finite-dimensional space.

Geometric integration: preserving structure

The most recent turn is structure-preserving geometric integration, which began in the 1980s and 1990s. The insight is that many ODEs come from physics and carry geometric structure—energy, symplectic form, momentum, time-reversibility—and that a numerical method which ignores this structure will drift, even if it is accurate order by order. A symplectic integrator, such as the Störmer–Verlet method for Hamiltonian systems, preserves the symplectic form exactly and conserves energy approximately over very long times, whereas a more accurate non-symplectic method may slowly spiral away1213. The theory was developed by Ruth, Feng Kang, Sanz-Serna, Hairer, Lubich, and others, and it has changed how long-time simulations in celestial mechanics, molecular dynamics, and plasma physics are done. The guiding principle—discretize the structure, not just the equation—is now a standard part of numerical analysis.

The present

Ordinary differential equations today is a field with several living strands. The classical theory of existence, uniqueness, and linear systems is taught everywhere and remains the foundation. Qualitative and dynamical-systems methods—phase portraits, stability, bifurcation, chaos—are the main tools for understanding nonlinear models41411. Numerical analysis supplies the solvers, with stiff and structure-preserving methods as mature specialties7915. Functional analysis provides the abstract framework, especially for linear and infinite-dimensional problems. The open questions are largely in the nonlinear and high-dimensional regime: rigorous understanding of chaotic systems, the design of integrators that preserve multiple structures at once, and the analysis of equations with non-smooth or random coefficients. The subject that began as a hunt for formulas is now, more than ever, the study of what solutions do when no formula exists.

References

  1. Ordinary differential equation (introduction, Solutions of finite duration, Linear differential equations), Wikipedia
  2. Differential equation (History), Wikipedia
  3. Linear differential equation, Wikipedia
  4. Qualitative theory of differential equations, Wikipedia
  5. Lyapunov stability (introduction, History), Wikipedia
  6. Stability theory, Wikipedia
  7. Numerical methods for ordinary differential equations (History, introduction), Wikipedia
  8. Runge–Kutta methods (introduction, Implicit Runge–Kutta methods), Wikipedia
  9. Stiff equation (introduction, Background), Wikipedia
  10. Dynamical system (History), Wikipedia
  11. Chaos theory, Wikipedia
  12. Symplectic integrator, Wikipedia
  13. Energy drift, Wikipedia
  14. Bifurcation theory, Wikipedia
  15. Geometric integrator, Wikipedia

The frameworks of Ordinary Differential Equations

Each framework is a way of working the field has used: when it held, and the question it answers.

  1. 1670-1830

    Solution by Quadrature

    How can a differential equation be solved by integrals?

    Fly to its tower in Nousfarer

  2. 1768-Present

    Numerical methods for ordinary differential equations

    How can a solution be computed approximately step by step?

    Fly to its tower in Nousfarer

  3. 1820-Present

    Existence and Uniqueness Theory

    When does an initial value problem have a unique solution?

    Fly to its tower in Nousfarer

  4. 1830-Present

    Linear Theory and Sturm–Liouville Spectral Theory

    How do linear equations decompose into independent solutions?

    Fly to its tower in Nousfarer

  5. 1880-Present
  6. 1892-Present

    Lyapunov stability

    When does a solution stay near equilibrium?

    Fly to its tower in Nousfarer

  7. 1948-Present

    Operator Semigroup Theory

    How can evolution equations be solved by operator families?

    Fly to its tower in Nousfarer

  8. 1950-Present

    Stiff Equations and Implicit Methods

    How can equations with widely separated time scales be integrated?

    Fly to its tower in Nousfarer

  9. 1960-Present

    Dynamical systems theory

    What is the long-term behaviour of a flow?

    Fly to its tower in Nousfarer

  10. 1980-Present

    Geometric integrator

    How can numerical methods preserve the geometry of the flow?

    Fly to its tower in Nousfarer

How the ideas connect

Each arrow runs from an earlier idea to the later one it shaped. The list below says what passed between them.

Solution by QuadratureNumerical methods for ordinary differ…Existence and Uniqueness TheoryLinear Theory and Sturm–Liouville Spe…Qualitative Theory of Differential Eq…Lyapunov stabilityOperator Semigroup TheoryStiff Equations and Implicit MethodsDynamical systems theoryGeometric integrator

Beyond the field

Where Ordinary Differential Equations's ideas came from in other fields, and what they led to there: the strongest 12 of 62 links.

A city in Nousfarer: a red-carpeted stair climbing terraces to a crowned tower on an acropolis, more towers round it, forest and snowy mountains beyond.
A city in Nousfarer, the game played in the browser.

Ordinary Differential Equations is a planet in Nousfarer

Its city has a tower for each of these 10 frameworks. Land, walk in, and climb one concept at a time.

Land on Ordinary Differential Equations

Related fields

Fields its ideas are linked to

More in Analysis

See where Ordinary Differential Equations sits on the Map of Human Knowledge