Initial-Value and Boundary-Value Problems for PDEs

1Introduction

A PDE problem is not determined by the equation alone. It is specified only after combining the domain of the unknown with the locations and kinds of prescribed data.

This lecture distinguishes initial-value, boundary-value, and initial-boundary-value problems and explains how to assess existence, uniqueness, stability, and compatibility rather than merely counting conditions.

2Sets on which the problem is posed

Let $$\mathrm{\Omega }\subset R^n$$\Omega\subset\mathbb R^n be a connected spatial domain, let $$\mathit{\partial }\mathrm{\Omega }$$\partial\Omega be its boundary, and let $$T>0$$T>0. Throughout this lecture, assume that the boundary is sufficiently smooth for an outward unit normal and the divergence theorem to be available. For time evolution, consider $$u(x,t)$$u(x,t) on the spacetime domain

The section $$\mathrm{\Omega }\times \left\{0\right\}$$\Omega\times\{0\} is the initial surface, while $$\mathit{\partial }\mathrm{\Omega }\times (0,T)$$\partial\Omega\times(0,T) is the lateral boundary. Initial and boundary conditions are prescribed on different sets.

3Three problem formulations

An {initial-value problem} prescribes data on an initial surface and asks for the subsequent evolution. More generally, a Cauchy problem prescribes data on an initial hypersurface; this track uses the standard example of data on $$t=0$$t=0 in the whole space $$R^n$$\mathbb R^n.

A {boundary-value problem} prescribes data on the boundary of a domain. The principal static examples here are Laplace's and Poisson's equations.

An {initial-boundary-value problem} concerns time evolution in a spatial domain with boundary and prescribes data on both the initial surface and the lateral boundary. Standard examples are the heat and wave equations on a bounded interval.

4Three principal boundary conditions

Let $$n$$n be the outward unit normal on $$\mathit{\partial }\mathrm{\Omega }$$\partial\Omega and define

The principal boundary conditions are:

• Dirichlet: $$u=g$$u=g, prescribing the value of the unknown on the boundary;
• Neumann: $${\mathit{\partial }}_{n}u=h$$\partial_nu=h, prescribing its outward normal derivative;
• Robin: $$\alpha u+\beta {\mathit{\partial }}_{n}u=r$$\alpha u+\beta\partial_nu=r, prescribing a linear combination of the value and normal derivative, with $$\alpha ,\beta$$\alpha,\beta not simultaneously zero.

Writing a Robin condition does not by itself imply well-posedness. Uniqueness and stability may require sign assumptions on the coefficients and conditions on the boundary decomposition.

The same condition need not be used on the entire boundary. If disjoint boundary portions satisfy $$\mathit{\partial }\mathrm{\Omega }={\mathrm{\Gamma }}_D\cup {\mathrm{\Gamma }}_N$$\partial\Omega=\Gamma_D\cup\Gamma_N, imposing Dirichlet data on $${\mathrm{\Gamma }}_D$$\Gamma_D and Neumann data on $${\mathrm{\Gamma }}_N$$\Gamma_N gives a mixed boundary condition. This differs from a Robin condition, which combines a value and a normal derivative at the same location.

The sign convention matters when Neumann data are interpreted as flux. If $$k>0$$k>0 is the thermal conductivity and $$q=-k\nabla u$$q=-k\nabla u is the heat flux, then the outward flux is

Thus $${\mathit{\partial }}_{n}u>0$$\partial_nu>0 does not mean that the outward heat flux is positive. The equation must be inspected to determine which quantity is prescribed.

5Time order and initial conditions

For standard evolution equations, the order in time indicates the number of initial data. The heat equation

is first order in time and is supplied with an initial distribution $$u(x,0)=f\left(x\right)$$u(x,0)=f(x). The wave equation

is second order in time and is supplied with initial displacement and velocity,

This is not a universal rule that exactly as many arbitrary conditions as the time order may always be imposed. Constraints in the equation or a characteristic data surface can change which data are independent. The next lecture treats the admissible data curves for first-order PDEs.

6Transport equation: selecting the boundary carrying data

Let $$c>0$$c>0 and consider on $$0<x<L$$0<x<L

Information travels to the right with speed $$c$$c. Hence $$x=0$$x=0 is the inflow boundary and $$x=L$$x=L is the outflow boundary. In addition to $$u(x,0)=f\left(x\right)$$u(x,0)=f(x), one ordinarily prescribes the inflow value $$u(0,t)=a\left(t\right)$$u(0,t)=a(t). The value $$u(L,t)$$u(L,t) is transported from the interior, so prescribing an arbitrary value there as well generally overdetermines the problem.

At the corner $$(0,0)$$(0,0), compatibility requires $$f\left(0\right)=a\left(0\right)$$f(0)=a(0). If $$c<0$$c<0, the inflow boundary changes to $$x=L$$x=L. Thus the location and number of boundary conditions are determined by the direction in which characteristics carry information, not merely by the number of geometric endpoints.

7Well-posedness

A {well-posed problem} satisfies the following three conditions in the specified solution class:

1. existence: at least one solution satisfies the conditions;
2. uniqueness: at most one solution satisfies them;
3. stability: small changes in the data produce small changes in the solution in the selected norms.

The third condition is continuous dependence in Hadamard's sense. The existence of an explicit formula alone does not prove well-posedness. The function spaces and norms used to measure data and solutions are part of the problem formulation. This lecture organizes the placement of conditions with classical solutions in mind; a later lecture uses energy estimates for uniqueness and continuous dependence.

In the backward heat problem, small errors in rapidly oscillating spatial components are amplified exponentially when reconstructing an earlier temperature. Even if the equation can be inverted formally, loss of continuous dependence makes the problem ill-posed. This illustrates why existence, uniqueness, and stability must be assessed separately.

8Compatibility of the initial surface and boundary

The initial surface and lateral boundary meet on $$\mathit{\partial }\mathrm{\Omega }\times \left\{0\right\}$$\partial\Omega\times\{0\}. Conditions preventing the two data from contradicting each other there are called compatibility conditions.

For the heat equation with

a classical solution continuous up to the closed domain must satisfy at least

For Neumann data $${\mathit{\partial }}_{n}u=h$$\partial_nu=h and a sufficiently smooth initial value, one needs $${\mathit{\partial }}_{n}f=h(·,0)$$\partial_nf=h(\cdot,0). Higher regularity requires higher-order compatibility conditions obtained using the equation.

For the one-dimensional wave equation with fixed endpoints $$u(0,t)=u(L,t)=0$$u(0,t)=u(L,t)=0, a smooth classical solution requires

If the boundary condition may be differentiated in time at the required regularity, then one also needs

Failure of compatibility does not always imply that no solution exists: weak solutions or solutions restricted to $$t>0$$t>0 may still be meaningful. The solution concept and required boundary continuity must be fixed first.

9Heat equation: the boundary changes the total quantity

Let $$u_t=\kappa \mathrm{\Delta }u$$u_t=\kappa\Delta u on a smooth bounded domain. The divergence theorem gives

Under homogeneous Neumann data $${\mathit{\partial }}_{n}u=0$$\partial_nu=0, the total quantity $${\int }_{\mathrm{\Omega }}udx$$\int_\Omega u\,dx is conserved. Under Dirichlet data, the boundary flux is generally nonzero, so the total quantity need not be conserved. Boundary conditions therefore describe interaction with the exterior rather than being merely computational accessories.

10Laplace and Poisson equations: the pure Neumann exception

Under suitable assumptions, boundary values in a Dirichlet problem determine a harmonic function uniquely. In a pure Neumann problem, however, if $$u$$u is a solution, then $$u+C$$u+C has the same normal derivative, so the solution is determined only up to a constant.

Integrating

and applying the divergence theorem gives the necessary condition

For Laplace's equation, $$F=0$$F=0, so $${\int }_{\mathit{\partial }\mathrm{\Omega }}hdS=0$$\int_{\partial\Omega}h\,dS=0 is necessary. If it fails, no solution exists. Even when it holds, uniqueness requires a normalization such as $${\int }_{\mathrm{\Omega }}udx=0$$\int_\Omega u\,dx=0.

11Reading conditions without excess or deficiency

• Fix the equation, spatial domain, time interval, and solution concept first.
• Distinguish initial-value, boundary-value, and initial-boundary-value problems.
• Check the outward-normal and flux sign conventions.
• Do not determine the number of conditions mechanically from the time order; inspect characteristics and constraints.
• Check compatibility where the initial surface meets the lateral boundary.
• Treat existence, uniqueness, and stability as separate claims.
• For a pure Neumann problem, check the integral condition and indeterminacy by constants.

12Scope of validity

This lecture focused on smooth domains and standard linear PDEs. Nonsmooth boundaries, discontinuous coefficients, and shocks in nonlinear equations require weak solutions and trace theory rather than only classical solutions. Proving well-posedness also requires tools appropriate to the equation's type, such as maximum principles, energy estimates, and eigenfunction expansions.

13Subsequent lectures

For first-order PDEs, the method of characteristics determines which curves can carry initial data. For second-order PDEs, classify the principal part and then examine natural problem formulations and information propagation under appropriate additional assumptions. The model comparison then examines how conditions affect heat, wave, and Laplace equations.