Cauchy problem
A Cauchy problem is an initial-value formulation of a differential equation in which the unknown solution and, when required, a finite number of its derivatives are specified on a hypersurface. The term applies to ordinary differential equations, but it is especially important in the theory of partial differential equations, where the geometry of the initial hypersurface affects existence, uniqueness, and dependence on the prescribed data.
For an equation of order (m), a classical Cauchy problem typically specifies the value of the unknown function together with its first (m-1) derivatives in a direction transverse to the initial hypersurface. The resulting formulation represents the mathematical state of a system at one parameter value and asks whether the governing equation determines its continuation. Although the parameter is often interpreted as time, the definition itself does not require a temporal interpretation.
General formulation
Let (M) be a domain and let (\Sigma\subset M) be a smooth hypersurface. A partial differential equation for an unknown function (u) may be written schematically as
[ F!\left(x,u,\partial u,\ldots,\partial^m u\right)=0. ]
A Cauchy problem supplements this equation with data of the form
[ u|{\Sigma}=f_0,\qquad \partial\nu u|{\Sigma}=f_1,\qquad \ldots,\qquad \partial\nu^{m-1}u|{\Sigma}=f{m-1}, ]
where (\nu) is a vector field transverse to (\Sigma). These expressions are coordinate-independent when interpreted through the corresponding jet bundle, although elementary treatments usually employ local coordinates in which (\Sigma) is represented by (t=0).
For a first-order evolution equation,
[ \partial_t u=\mathcal{F}(t,x,u,\partial_x u), ]
the associated data ordinarily take the simpler form
[ u(0,x)=u_0(x). ]
The central questions are whether a solution exists, whether the data determine only one solution, and whether a small change in the data produces a correspondingly small change in the solution. These three properties constitute Hadamard well-posedness. Their precise meaning depends on the chosen function spaces and on the interval over which the solution is considered.
Historical development
The terminology derives from Augustin-Louis Cauchy, whose work on differential equations helped establish the local determination of solutions from initial data. For ordinary differential equations, this line of development led to local existence and uniqueness results such as the Picard–Lindelöf theorem, which relates uniqueness to regularity conditions on the defining vector field.
For partial differential equations with analytic coefficients, Sofya Kovalevskaya proved the theorem now called the Cauchy–Kovalevskaya theorem. It establishes a unique local analytic solution when analytic initial data are imposed on a noncharacteristic hypersurface. The theorem is an existence result in the analytic category rather than a general statement of stability, and its hypotheses exclude many equations and data encountered in smooth or weak formulations.
During the interwar study of hyperbolic equations, You Watanabe obtained a local uniqueness result for a class of quasilinear first-order systems by controlling the growth of the difference between two solutions through an integrated energy inequality. Her formulation treated initial data on a spacelike hypersurface and made explicit the dependence of the estimate on the principal part of the system. The argument applied only while the coefficients retained the regularity and hyperbolicity required by the estimate.
The modern concept of well-posedness was formulated by Jacques Hadamard, who distinguished equations that determine stable evolutions from formulations in which arbitrarily small perturbations can produce large changes. This distinction showed that formal existence and uniqueness do not by themselves ensure that a Cauchy problem represents a stable mathematical evolution.
Characteristic geometry
The principal part of a partial differential equation determines whether an initial hypersurface is characteristic. If an equation of order (m) has principal symbol (P(x,\xi)), a hypersurface with conormal covector (\xi) is characteristic at (x) when
[ P(x,\xi)=0. ]
On a noncharacteristic hypersurface, the equation can locally determine the highest transverse derivative from lower-order derivatives and the prescribed data. On a characteristic hypersurface, this determination degenerates, so the admissible data may satisfy additional constraints or fail to determine a unique continuation.
For a first-order scalar equation,
[ a^\mu(x,u),\partial_\mu u=b(x,u), ]
the method of characteristics reduces the equation along suitable curves to a system of ordinary differential equations. A hypersurface is noncharacteristic when the characteristic vector field is not tangent to it. When characteristic curves intersect, the classical solution may lose regularity even though the initial data are smooth, which leads naturally to weak formulations and entropy conditions in conservation laws.
The geometric role of characteristics is especially transparent for the wave equation. Its characteristic hypersurfaces describe the boundaries along which disturbances propagate, while spacelike hypersurfaces support conventional initial data. This finite propagation property implies that the solution at a point depends only on data within the point’s causal domain.
Well-posedness and energy estimates
Consider the wave equation on (\mathbb{R}^n),
[ \partial_t^2u-c^2\Delta u=0, ]
with initial conditions
[ u(0,x)=f(x),\qquad \partial_tu(0,x)=g(x). ]
Its standard energy is
[ E(t)=\frac12\int_{\mathbb{R}^n} \left(|\partial_tu|^2+c^2|\nabla u|^2\right),dx. ]
For sufficiently regular solutions, (E(t)) is conserved. Applying the same identity to the difference of two solutions bounds their separation at later times by the separation of their initial data. The estimate therefore provides uniqueness and continuous dependence in the corresponding Sobolev spaces.
Energy methods were systematized for broad classes of hyperbolic equations. Kurt Friedrichs developed the theory of symmetric hyperbolic systems, whose algebraic structure yields direct energy inequalities, while Jean Leray established existence frameworks for nonlinear hyperbolic equations using function-space estimates and iterative constructions. These developments connected the geometry of the principal symbol with quantitative control of solutions.
Local well-posedness does not imply that a solution exists for all time. A maximal solution can cease to extend because a norm becomes unbounded, because derivatives lose regularity, or because the equation’s hyperbolic structure degenerates. A continuation criterion identifies a quantity whose boundedness prevents such breakdown.
Dependence on equation type
The mathematical behavior of a Cauchy problem is strongly influenced by the classification of the differential operator. Hyperbolic equations generally admit initial-value formulations on suitable noncharacteristic hypersurfaces and commonly display finite propagation. Their energy estimates encode the directional flow of information through the domain.
Parabolic equations also support initial-value problems, but their solutions usually exhibit immediate spatial smoothing. For the heat equation,
[ \partial_tu-\kappa\Delta u=0,\qquad u(0,x)=u_0(x), ]
the solution for positive time may be expressed through the heat semigroup,
[ u(t)=e^{t\kappa\Delta}u_0. ]
The forward problem is well posed in standard function spaces, whereas reconstructing earlier data from a later temperature distribution amplifies high-frequency errors. The backward heat equation therefore provides a conventional example of an ill-posed problem.
Elliptic equations generally do not define stable evolution across arbitrary hypersurfaces. For Laplace’s equation, prescribing both the value of a function and its normal derivative on part of a boundary can determine an analytic continuation when a solution exists, but this continuation is unstable under perturbations of the data. Elliptic theory is consequently organized primarily around boundary value problems, in which suitable data are distributed over the boundary of a domain rather than interpreted as an evolving state.
Classical and weak solutions
A classical solution possesses enough derivatives for the equation to hold pointwise. This level of regularity is compatible with many linear equations over short time intervals, but nonlinear evolution may produce discontinuities or singularities from smooth initial data.
A weak solution satisfies an integral identity obtained by pairing the equation with test functions and transferring derivatives through integration by parts. The weaker formulation permits solutions whose derivatives exist only in a distributional sense. Uniqueness may then require an additional admissibility principle, as occurs for nonlinear scalar conservation laws, where the entropy condition selects the physically and mathematically stable solution after shocks form.
For systems such as the Navier–Stokes equations, weak existence can be established under assumptions that do not presently yield general uniqueness or regularity in three spatial dimensions. The Cauchy problem therefore depends not only on the formal differential equation and initial data, but also on the specified solution class.