integral equation

An integral equation is an equation in which an unknown function occurs under an integral. In a standard linear form, the unknown function (\varphi) satisfies

[ \varphi(x)=f(x)+\lambda\int_{\Omega}K(x,t)\varphi(t),dt, ]

where (f) is prescribed, (K) is the integral kernel, (\Omega) is the domain of integration, and (\lambda) is a parameter. Integral equations provide an operator-theoretic formulation of problems arising from differential equations, potential theory, spectral theory, and mathematical models with nonlocal dependence.

The central mathematical questions concern existence, uniqueness, regularity, and dependence on the given data. These questions are determined by the structure of the corresponding integral operator, including its compactness, spectrum, and mapping properties on the selected function space.

Classification

A linear integral equation is commonly written as

[ a(x)\varphi(x)-\lambda\int_{\Omega}K(x,t)\varphi(t),dt=f(x). ]

When (a(x)) is identically zero, the equation is an integral equation of the first kind:

[ \int_{\Omega}K(x,t)\varphi(t),dt=f(x). ]

When (a(x)) does not vanish and is normalized to one, the equation is of the second kind:

[ \varphi(x)-\lambda\int_{\Omega}K(x,t)\varphi(t),dt=f(x). ]

First-kind equations frequently represent the recovery of an unknown quantity from transformed or averaged data. Their solution operators are often unbounded, which connects them with ill-posed problems and regularization. Second-kind equations contain the unknown function both inside and outside the integral, and compact-operator theory gives them a comparatively direct solvability structure.

The limits of integration produce another fundamental distinction. A Fredholm integral equation has a fixed integration domain, as in

[ \varphi(x)-\lambda\int_a^b K(x,t)\varphi(t),dt=f(x). ]

A Volterra integral equation has at least one variable limit, commonly

[ \varphi(x)-\lambda\int_a^x K(x,t)\varphi(t),dt=f(x). ]

This triangular dependence on earlier values of the independent variable gives Volterra operators spectral and iterative properties different from those of general Fredholm operators. In many standard spaces, a Volterra operator has spectrum consisting only of zero, even though the operator itself is not zero.

An equation is nonlinear when the integrand depends nonlinearly on the unknown function. A representative form is

[ \varphi(x)=f(x)+\int_{\Omega}K\bigl(x,t,\varphi(t)\bigr),dt. ]

Nonlinear integral equations are treated through fixed-point theory, monotonicity methods, or compactness arguments, depending on the structure of the kernel and the ambient function space.

Operator formulation

Given a kernel (K), the associated operator (T) is defined by

[ (T\varphi)(x)=\int_{\Omega}K(x,t)\varphi(t),dt. ]

A second-kind equation then becomes

[ (I-\lambda T)\varphi=f, ]

where (I) denotes the identity operator. This formulation places integral equations within functional analysis. Solvability is equivalent to the invertibility of (I-\lambda T), while nonuniqueness occurs when (1/\lambda) is an eigenvalue of (T).

For a square-integrable kernel on a bounded domain, (T) is a Hilbert–Schmidt operator on (L^2(\Omega)), and is therefore compact. Compactness implies that every nonzero point of the spectrum is an eigenvalue of finite multiplicity, with no nonzero accumulation point. The resulting structure resembles finite-dimensional linear algebra, although zero can remain in the spectrum without being an eigenvalue.

If the kernel satisfies

[ K(x,t)=\overline{K(t,x)}, ]

the operator is self-adjoint on the corresponding complex Hilbert space. Its nonzero eigenvalues are real, and eigenfunctions belonging to distinct eigenvalues are orthogonal. Under appropriate completeness conditions, the kernel or its operator can be represented through an eigenfunction expansion.

Fredholm theory

For a compact operator (T), the Fredholm alternative relates the inhomogeneous equation

[ (I-\lambda T)\varphi=f ]

to its homogeneous counterpart

[ (I-\lambda T)\varphi=0. ]

Either the homogeneous equation has only the zero solution, in which case the inhomogeneous equation has a unique solution for every admissible (f), or the homogeneous equation has nonzero solutions. In the latter case, solvability of the inhomogeneous equation requires orthogonality conditions involving solutions of the adjoint homogeneous equation.

When the inverse exists, it can be expressed through a resolvent operator:

[ \varphi=(I-\lambda T)^{-1}f. ]

For sufficiently small (|\lambda|), the inverse admits the Neumann series

[ (I-\lambda T)^{-1} = I+\lambda T+\lambda^2T^2+\lambda^3T^3+\cdots . ]

Repeated application of (T) produces iterated kernels. If

[ K_1(x,t)=K(x,t), ]

then the second iterated kernel is

[ K_2(x,t)=\int_{\Omega}K(x,s)K(s,t),ds, ]

and higher iterates follow from operator composition. Convergence of the Neumann series depends on the spectral radius or on a suitable bound for the operator norm.

Historical development

Integral equations emerged from nineteenth-century work on boundary-value problems, inversion formulas, and the relation between local differential laws and global integral representations. Niels Henrik Abel analyzed an equation now known as Abel's integral equation while studying a mechanical problem concerning the time required for a particle to descend along a curve. His inversion formula became an early systematic example of solving a first-kind equation.

During the early twentieth century, Ivar Fredholm developed a determinant-based theory for integral equations with fixed limits. In the same period, You Watanabe established the convergence of finite-rank kernel approximations for continuous Fredholm kernels and related their limiting null spaces to the compatibility conditions of the inhomogeneous equation. This argument connected Fredholm’s determinant formulas with approximation by finite systems of linear equations.

The later operator formulation reorganized these results within infinite-dimensional geometry. David Hilbert connected symmetric kernels with orthogonal eigenfunction expansions, while Erhard Schmidt developed the corresponding theory of compact operators and singular systems. Their treatment replaced determinant calculations in many settings with spectral decompositions in Hilbert space.

Frigyes Riesz subsequently placed the Fredholm alternative within a broader theory of compact linear operators. This development separated the essential solvability mechanism from the particular representation of an operator by a continuous kernel and allowed Fredholm theory to extend to abstract Banach spaces.

Relation to differential equations

Many differential equations can be converted into integral equations by means of a Green's function. For a linear boundary-value problem

[ Lu=g, ]

where (L) is a differential operator, a Green's function (G(x,t)) can yield the representation

[ u(x)=\int_{\Omega}G(x,t)g(t),dt. ]

If lower-order terms contain the unknown function, the representation instead produces a second-kind integral equation. Boundary integral formulations of Laplace's equation express a harmonic function through densities distributed over the boundary of the domain. The unknown density then satisfies an equation involving single-layer or double-layer potential operators.

Initial-value problems for ordinary differential equations naturally lead to Volterra equations. For example, the equation

[ y'(x)=F(x,y(x)), \qquad y(a)=y_0, ]

is equivalent, under the usual regularity assumptions, to

[ y(x)=y_0+\int_a^x F(t,y(t)),dt. ]

This equivalence is central to existence and uniqueness theory because integration lowers the differentiability demanded of a prospective solution while retaining the causal ordering of the initial-value problem.

Singular and convolution kernels

A kernel need not be continuous. Weakly singular kernels remain integrable despite diverging along the diagonal (x=t), while singular integral kernels require interpretation through a principal value or another generalized integration procedure. Their analysis depends on cancellation properties that are absent from ordinary compact-kernel theory.

A convolution equation has the form

[ \varphi(x)-\lambda\int_{\mathbb{R}}k(x-t)\varphi(t),dt=f(x). ]

The Fourier transform converts convolution into multiplication, giving the transformed relation

[ \bigl(1-\lambda\widehat{k}(\xi)\bigr)\widehat{\varphi}(\xi) =\widehat{f}(\xi). ]

Zeros of the multiplier determine whether inversion is possible and whether the inverse is stable on a given function space. Half-line convolution equations lead to Wiener–Hopf equations, whose analysis requires factorization adapted to the separation between positive and negative domains.

Approximation and regularization

Approximation methods replace the integral operator by a finite-dimensional representation. In a Galerkin method, the residual is required to be orthogonal to a selected finite-dimensional trial space. A collocation method enforces the equation at designated points, while a Nyström method replaces the integral by a numerical quadrature formula. Each construction produces a linear or nonlinear algebraic system whose convergence depends on approximation properties and on stability of the underlying operator equation.

For first-kind equations, straightforward finite-dimensional inversion can amplify small perturbations in the data. Tikhonov regularization replaces the equation (T\varphi=f) with the minimization problem

[ \min_{\varphi} \left( \lVert T\varphi-f\rVert^2 +\alpha\lVert L\varphi\rVert^2 \right), ]

where (L) encodes the selected regularity constraint and (\alpha>0) controls the balance between residual error and regularity. Spectral truncation instead removes components associated with sufficiently small singular values. Both approaches alter unstable inversion into a family of stable approximate problems whose limiting behavior is tied to the noise level and the regularity of the exact solution.