Loading field map
The central tension that drives the study of partial differential equations (PDEs) is deceptively simple: given an equation linking a function of several variables to its partial derivatives, can one determine whether a solution exists, whether it is unique, and how its properties depend on the data? For equations that model physical phenomena—heat diffusion, wave propagation, fluid flow, or the shape of a soap film—these questions are not merely mathematical curiosities. They determine whether a model is predictive, whether small changes in initial conditions produce wild swings in outcomes, and whether the solution remains smooth or develops singularities. The history of PDE theory is the history of learning to answer these questions as the equations themselves grew more complex and less forgiving.
The first sustained attempts to solve PDEs relied on explicit formulas. For the wave equation, d'Alembert discovered in 1747 that the general solution could be written as a sum of two traveling waves, an approach later systematized as the method of characteristics. For the heat equation, Fourier introduced separation of variables and the representation of solutions as infinite series of trigonometric functions—Fourier series—a technique that forced mathematicians to confront what it meant for a function to be represented by a series at all. Laplace's equation, governing steady-state phenomena, yielded to similar separation-of-variables strategies in suitable coordinate systems, and Green's functions provided an integral representation for solutions on bounded domains.
By the end of the nineteenth century, these scattered techniques had been organized into a classification system that remains a first reference today. Equations are typed as elliptic (Laplace's equation), parabolic (the heat equation), or hyperbolic (the wave equation), each type exhibiting characteristic behavior: elliptic equations smooth out data and require boundary conditions, parabolic equations describe irreversible evolution toward equilibrium, and hyperbolic equations propagate disturbances at finite speed. Jacques Hadamard crystallized the expectations for any well-posed problem: a solution should exist, be unique, and depend continuously on the data. The classical framework treated a solution as a function with enough continuous derivatives to satisfy the equation pointwise, and its methods succeeded brilliantly for linear equations with simple geometries. But the moment the domain became irregular, the coefficients became discontinuous, or the equation became nonlinear, the classical toolkit reached its limit. The method of characteristics could not handle shocks; separation of variables failed on arbitrary domains; and pointwise differentiability was too rigid a requirement for many physically meaningful problems.
The breakthrough that extended PDE theory beyond the classical limitations came from importing the language of functional analysis. Instead of seeking a smooth function that satisfies the equation at every point, one could look for a function that satisfies the equation in an integrated sense after multiplying by a smooth test function and integrating by parts. This idea—the weak solution—shifted the problem from pointwise identities to integral identities, and it required a new space of functions large enough to contain these weak solutions yet structured enough to recover classical regularity when the data permitted.
Sobolev spaces provided exactly that habitat. A function belongs to a Sobolev space if it and its distributional derivatives (in a sense made precise by Laurent Schwartz's theory of distributions) lie in an Lp space. Integration by parts, the workhorse of the weak formulation, becomes a definitional property rather than a derived one. The Lax–Milgram theorem, a variant of the Riesz representation theorem, gave a clean existence and uniqueness result for elliptic equations in weak form, and the Fredholm alternative extended this to equations that are not coercive. For parabolic and hyperbolic equations, semigroup theory—developed by Einar Hille, Kōsaku Yosida, and others—treated the time variable as an evolution in a Banach space, linking PDE theory to operator theory and spectral theory.
This functional analytic framework absorbed the classical classification rather than rejecting it. Elliptic regularity theory showed that weak solutions of elliptic equations with smooth coefficients are actually smooth, vindicating the classical search for pointwise solutions in the cases where they exist. The framework also handled problems that classical methods could not touch: equations with rough coefficients, domains with corners, and boundary conditions understood only in a weak sense. But the framework was built primarily for linear equations. When nonlinear terms appeared, the compactness arguments that worked for linear problems often failed, and the linear functional analytic machinery could not directly produce solutions.
The turn to nonlinear problems forced a diversification of methods. No single technique could cover the range of nonlinear phenomena—shock waves, free boundaries, minimal surfaces, nonlinear dispersion, and kinetic equations—so the field fragmented into specialized toolkits, each adapted to a particular type of nonlinearity.
For first-order conservation laws, which model traffic flow and gas dynamics, solutions develop discontinuities (shocks) even from smooth initial data. Classical pointwise solutions cease to exist after the shock forms, and weak solutions are not unique without an additional selection criterion. The theory of viscosity solutions, developed by Pierre-Louis Lions and Michael Crandall in the 1980s, provided a notion of solution for Hamilton–Jacobi equations and, later, for fully nonlinear second-order equations. A viscosity solution is defined by comparing it with smooth test functions that touch it from above or below; the definition is entirely nonvariational and does not rely on integration by parts, making it suitable for equations where weak solutions in the Sobolev sense are not available.
For systems of conservation laws in several space dimensions, the method of compensated compactness, introduced by Luc Tartar and Ronald DiPerna, used weak convergence and the structure of the nonlinearity to extract a solution from a sequence of approximate solutions. This technique revealed that the nonlinearity itself could enforce compactness that linear theory could not provide.
Geometric PDEs—equations like the minimal surface equation, the mean curvature flow, and the Yamabe equation—drew on the calculus of variations and geometric measure theory. The direct method of the calculus of variations, which seeks a minimizer of an energy functional, often produces a weak solution that is not smooth enough to satisfy the Euler–Lagrange equation in the classical sense. Geometric measure theory, with its tools of rectifiable sets, currents, and varifolds, provided the infrastructure to study minimal surfaces and free boundary problems where the solution itself is a geometric object rather than a function.
Nonlinear dispersive equations, such as the nonlinear Schrödinger equation and the Korteweg–de Vries equation, required harmonic analysis, Strichartz estimates, and the method of bilinear estimates to prove local and global well-posedness. These techniques exploit the cancellation between different frequencies, a phenomenon invisible to the purely functional analytic approach.
Kinetic equations, including the Boltzmann equation, demanded a combination of averaging lemmas, velocity averaging, and renormalized solutions—a notion introduced by DiPerna and Lions that allowed solutions to be defined even when the nonlinearity involves products of functions that are not integrable.
Today, the three frameworks coexist, each with its own domain of strength. Classical PDE theory remains the default for teaching and for problems where explicit formulas are available. Functional analytic methods are the workhorse for linear and semilinear elliptic, parabolic, and hyperbolic equations, and they provide the regularity theory that underpins numerical analysis. Modern nonlinear methods are essential whenever the nonlinearity is strong enough to break the linear functional analytic machinery.
What the leading frameworks agree on is the primacy of well-posedness as a guiding principle: a good theory should produce existence, uniqueness, and continuous dependence on data, even if the notion of solution must be adapted to the equation. They also agree that regularity—the recovery of smoothness from weak solutions—is a central goal, and that the classification by type (elliptic, parabolic, hyperbolic) remains useful even for nonlinear equations.
Where they disagree is on the role of linearization. One tradition, rooted in the functional analytic framework, treats nonlinear problems as perturbations of linear ones, using implicit function theorems, fixed-point arguments, and linearized stability analysis. Another tradition, emerging from the modern nonlinear framework, insists that many nonlinear phenomena are irreducibly nonlinear: shocks, singularities, and pattern formation cannot be captured by linearization, and the methods must be adapted to the specific structure of the nonlinearity. This tension is not a sign of weakness but a productive division of labor. The most active frontiers—free boundary problems, the analysis of the Navier–Stokes equations, the study of nonlinear waves, and the rigorous derivation of macroscopic equations from microscopic models—draw on all three frameworks, combining functional analytic estimates with nonlinear techniques tailored to the problem at hand.
The trajectory of PDE theory reveals a field that has repeatedly expanded its notion of what counts as a solution, each time absorbing the previous framework while opening new questions that the old tools could not address. The classical search for explicit formulas gave way to the functional analytic search for weak solutions, which in turn gave way to a pluralistic landscape of specialized notions of solution, each designed to capture the essential structure of a particular nonlinearity. The field's unity lies not in a single method but in the persistent questions: existence, uniqueness, regularity, and the relationship between the equation and the phenomena it describes.