Loading field map
A partial differential equation (PDE) is an equation that relates a function of several variables to its own partial derivatives. Where an ordinary differential equation (ODE) describes a system evolving along a single independent variable—typically time—a PDE describes a quantity that varies over a continuum of positions, or over both space and time. The unknown is not a number or a curve, but a field: a temperature distribution, a velocity profile, a stress pattern, or a quantum wavefunction. PDEs are the language in which most of classical and modern physics, engineering science, and increasingly biology and finance, is written.
The central problem of the subfield is not merely to write down such equations, but to understand what their solutions look like, whether they exist, whether they are unique, and how they depend on the data that define them. Because most PDEs cannot be solved by a closed formula, the discipline is built around a hierarchy of strategies: exact methods for special classes, qualitative analysis that extracts behavior without solving, and numerical algorithms that approximate solutions on a computer. The field is defined less by a single method than by a persistent tension between the desire for general theorems and the need for concrete, computable answers.
The first step in navigating PDEs is classification, because the type of an equation determines almost everything about its behavior. The most important distinction is among three types that arise in the classical physics of continua.
Elliptic equations describe equilibrium states. The prototypical example is Laplace's equation, ∇²$u = 0$, whose solutions are called harmonic functions. It governs steady-state temperature distributions, electrostatic potentials, and the shapes of minimal surfaces. The defining feature of elliptic problems is that information propagates in all directions: the value of the solution at any interior point depends on the entire boundary. These problems are typically posed on a bounded region with conditions specified on the boundary, and they are well-posed in the sense that small changes in the boundary data produce small changes in the solution.
Parabolic equations describe diffusion and relaxation toward equilibrium. The heat equation, ∂u/∂t = ∇²u, is the canonical example. It models how temperature spreads through a medium, how ink disperses in water, and how option prices evolve in the Black–Scholes model. Parabolic equations smooth out irregularities: even if the initial data are rough, the solution becomes infinitely smooth for any positive time. They have a directional character—time moves forward—but the spatial part behaves elliptically at each instant.
Hyperbolic equations describe wave propagation and transport. The wave equation, ∂²u/∂t² = c²∇²u, governs vibrations of strings, membranes, and electromagnetic fields. Hyperbolic equations preserve features: a sharp pulse travels without smoothing, and information propagates along characteristic curves at finite speed. They are the most delicate to analyze because solutions can develop discontinuities, as in shock waves in gas dynamics.
This threefold classification, formalized by the French mathematician Jacques Hadamard in the early twentieth century, is not a mere taxonomy. It determines which boundary conditions are admissible, whether solutions are smooth or singular, and which numerical methods will be stable. A problem that is well-posed for one type—say, specifying the value of the solution on the entire boundary for an elliptic equation—is ill-posed for another. The classification is local: an equation can change type in different regions of its domain, as happens in transonic flow, where the governing equation is elliptic in the subsonic region and hyperbolic in the supersonic region.
Hadamard's concept of well-posedness became the organizing principle of the field. A problem is well-posed if a solution exists, is unique, and depends continuously on the data. The third condition is crucial: if arbitrarily small changes in the initial or boundary data produce large changes in the solution, then the problem is useless for prediction, because real data are never known exactly.
This criterion was not merely a philosophical preference. Hadamard showed that certain problems that look reasonable—such as the Cauchy problem for Laplace's equation, where one specifies the value and normal derivative on a curve—are ill-posed: solutions may not exist, and when they do, they can change dramatically under tiny perturbations of the data. Such problems are now called improperly posed or inverse problems. They arise naturally in applications like medical imaging and geophysical exploration, where one tries to infer interior structure from boundary measurements. The theory of ill-posed problems is a substantial subfield in its own right, concerned with regularization: adding constraints or smoothing to make an unstable problem tractable.
Well-posedness is not guaranteed by the equation alone; it depends on the domain and the boundary conditions. The same elliptic operator can yield a well-posed Dirichlet problem (specifying the value on the boundary) and a subtly ill-posed Neumann problem (specifying the normal derivative) on certain domains. Understanding which combinations of equation, domain, and data yield well-posed problems is a large part of the subject.
Before the development of general theory, PDEs were studied through explicit formulas. The most powerful classical technique is separation of variables: one assumes the solution is a product of functions, each depending on a single variable, and reduces the PDE to a system of ODEs. This works when the domain has high symmetry—a rectangle, a circle, a sphere—and when the equation is linear with constant coefficients. The solutions are expressed as infinite series of special functions: Fourier series for periodic problems, Bessel functions for circular domains, Legendre polynomials for spherical ones.
The method of characteristics applies to first-order PDEs, which describe transport of quantities along curves. For a linear or quasilinear first-order equation, the solution is constant along certain curves called characteristics, and the problem reduces to solving ODEs along those curves. This method extends to some second-order hyperbolic equations, where characteristics are the paths along which information travels.
Integral transforms—Fourier and Laplace transforms—convert constant-coefficient PDEs into algebraic equations or ODEs in the transform variable. The heat equation, for instance, becomes an ordinary differential equation under the Fourier transform, and the solution is obtained by transforming back. These methods are powerful but limited: they require linearity, constant coefficients, and often infinite or highly symmetric domains.
The most important limitation of exact methods is that they fail for nonlinear equations, which are the rule in real applications. The Navier–Stokes equations of fluid dynamics, the Einstein field equations of general relativity, and the reaction–diffusion systems of biology are all nonlinear. For these, no general closed-form solutions exist, and the field must rely on qualitative and numerical approaches.
The twentieth century brought a fundamental shift in how PDEs are understood. The classical notion of a solution—a function with enough derivatives to plug into the equation—proved too restrictive. Many physical phenomena, such as shock waves in gas dynamics, produce solutions that are not differentiable, or even discontinuous. Yet these nonsmooth objects are exactly what the equations predict and what experiments observe.
The resolution came through the theory of distributions, developed by Laurent Schwartz in the 1940s. A distribution is a continuous linear functional on a space of smooth test functions; it generalizes the notion of a function so that every locally integrable function is a distribution, and so are derivatives of such functions, including the Dirac delta "function" that represents a point source. In this framework, one can define a weak solution of a PDE: a function that satisfies the equation when integrated against arbitrary smooth test functions. The derivatives are taken in the distributional sense, so the equation makes sense even when the solution is not differentiable.
This move transformed the field. It allowed mathematicians to prove existence of solutions for broad classes of nonlinear equations, because the space of distributions is large enough to contain limits of approximating sequences. The celebrated existence theory for the Navier–Stokes equations, due to Jean Leray in the 1930s, used weak solutions to show that solutions exist for all time in three dimensions, though their regularity remains an open question (one of the Clay Millennium Prize Problems).
The theory of weak solutions is intimately connected to the calculus of variations and Sobolev spaces—function spaces that measure not only the size of a function but also the size of its derivatives in an integrated sense. Many elliptic and parabolic equations can be reformulated as minimization problems: the solution is the function that minimizes an energy functional. This variational perspective provides both existence proofs and numerical methods, because minimizers can be approximated by finite-dimensional optimization.
Nonlinear PDEs are not merely harder versions of linear ones; they exhibit qualitatively new phenomena. The most striking is the formation of singularities: solutions that start smooth can develop discontinuities or blow up to infinity in finite time. Shock waves in compressible flow are the classic example. The equations of gas dynamics are hyperbolic, and their solutions develop discontinuities even from smooth initial data. Understanding these singularities requires a notion of weak solution that admits discontinuities, but weak solutions are not unique: the same initial data can produce many different weak solutions. The physical solution must be selected by an additional criterion, such as an entropy condition that encodes the second law of thermodynamics.
Other nonlinear phenomena include solitons—localized waves that travel without changing shape and pass through each other without interacting. These arise in integrable systems such as the Korteweg–de Vries equation for shallow water waves. The discovery of solitons in the 1960s led to the development of the inverse scattering transform, a method that solves certain nonlinear equations by a nonlinear analogue of the Fourier transform. This is a rare case where a nonlinear equation admits exact solutions, and it revealed deep connections to algebraic geometry and quantum field theory.
The theory of nonlinear PDEs is not a unified subject but a collection of techniques tailored to particular types of equations. Conservation laws, which express the balance of a quantity like mass or momentum, are studied through the method of characteristics and entropy conditions. Reaction–diffusion equations, which model pattern formation in biology and chemistry, are analyzed through stability theory and bifurcation analysis. Fully nonlinear equations, such as the Monge–Ampère equation of differential geometry, require sophisticated techniques from convex analysis and the theory of viscosity solutions.
Because most PDEs cannot be solved exactly, numerical approximation is not an afterthought but an integral part of the field. The two dominant approaches are finite difference methods and finite element methods.
Finite difference methods approximate the derivatives in the PDE by difference quotients on a grid. The PDE becomes a system of algebraic equations for the values of the solution at grid points. These methods are simple to implement and well understood for problems on regular domains, but they struggle with complex geometries and with conservation laws, where naive discretizations can produce spurious oscillations.
Finite element methods divide the domain into small pieces—triangles in two dimensions, tetrahedra in three—and approximate the solution by piecewise polynomial functions on each piece. The PDE is reformulated in weak form, and the approximate solution is chosen to satisfy the equation when tested against a finite-dimensional space of test functions. This approach handles complex geometries naturally and has a rigorous error theory rooted in the variational formulation. It is the method of choice for structural mechanics, electromagnetics, and many other engineering applications.
Spectral methods, which expand the solution in global basis functions such as Fourier modes or Chebyshev polynomials, achieve very high accuracy for smooth problems but are limited to simple domains. For hyperbolic problems with shocks, specialized methods such as Godunov schemes and essentially non-oscillatory (ENO) methods are designed to capture discontinuities without smearing them or introducing oscillations.
The numerical analysis of PDEs is concerned with two questions: stability (does the numerical solution remain bounded as the grid is refined?) and convergence (does the numerical solution approach the true solution?). The Lax equivalence theorem states that for linear problems, stability is equivalent to convergence, but for nonlinear problems the situation is more delicate. The field has developed a deep theory of error estimates, adaptive mesh refinement, and domain decomposition methods for parallel computing.
The present state of PDEs is characterized by a productive interplay among the analytical, variational, and computational traditions. The analytical tradition, rooted in the theory of weak solutions and a priori estimates, continues to address fundamental questions of existence, uniqueness, and regularity. The variational tradition, connected to the calculus of variations and geometric analysis, has produced deep results on minimal surfaces, Ricci flow, and other geometric PDEs. The computational tradition has grown enormously with the increase in computing power, and modern research often combines all three: numerical experiments suggest conjectures, analytical methods prove them, and variational principles guide the design of numerical schemes.
Several areas are particularly active. The regularity theory of elliptic and parabolic equations—understanding exactly how smooth solutions are and under what conditions singularities can occur—has seen major advances through the work of Luis Caffarelli and others. The study of free boundary problems, where the boundary of the region where the solution is nonzero is itself unknown, connects PDEs to probability theory and materials science. Kinetic equations, which describe the evolution of probability distributions in phase space, are central to plasma physics and rarefied gas dynamics. And the field of optimal transport, which seeks the most efficient way to move mass from one distribution to another, has revealed deep connections between PDEs, geometry, and economics.
The relationship between PDEs and probability has become especially fruitful. The Feynman–Kac formula represents solutions of certain parabolic equations as expectations over stochastic processes, and the theory of stochastic PDEs—equations driven by random noise—has grown into a major subfield with applications to statistical mechanics and finance. Conversely, PDE methods have been used to prove results in probability theory, such as the convergence of random walks to Brownian motion.
The open problems in the field remain formidable. The global regularity of the three-dimensional Navier–Stokes equations—whether smooth solutions can develop singularities in finite time—is perhaps the most famous. The question of whether solutions to certain nonlinear wave equations exist for all time, or blow up, is similarly unresolved in general. These are not merely technical difficulties; they reflect a genuine lack of understanding of how nonlinearity and geometry interact in the evolution of fields.
The field of PDEs is thus not a settled body of knowledge but a living discipline organized around a few central questions: What does it mean for a problem to be well-posed? How do singularities form and how can they be described? How can solutions be approximated reliably? The answers to these questions have changed dramatically over the past century, from the classical theory of explicit solutions to the modern theory of weak solutions and the contemporary synthesis of analysis, geometry, and computation. What remains constant is the conviction that the behavior of continuous systems—from the flow of air over a wing to the evolution of the universe—can be understood through the equations that govern them.