Chapter 7 — Linear Integral Equations
Equations in which the unknown function sits under an integral sign: how initial and boundary value problems become Volterra and Fredholm equations, how separable kernels reduce them to linear algebra, and how characteristic numbers, the Neumann series and the resolvent kernel describe every solution at once.
In 1823 the young Norwegian Niels Henrik Abel asked a question about a bead sliding without friction down a wire: if we know, for every starting height, how long the bead takes to reach the bottom, can we recover the shape of the wire? The unknown — the slope of the wire as a function of height — appears inside an integral, $\int_0^x \frac{\phi(t)}{\sqrt{x-t}}\,dt=f(x)$, and Abel solved this equation exactly. Seventy years later Vito Volterra met equations of the same kind when he modelled materials with memory and populations whose growth depends on their whole past: the present state is an integral over the history, $\int_0^x K(x,t)\,y(t)\,dt$. In 1900 Ivar Fredholm, studying the Dirichlet problem for Laplace's equation, treated equations whose integral runs over a fixed interval and discovered that they behave exactly like systems of linear equations: either there is exactly one solution for every right-hand side, or the homogeneous equation has nontrivial solutions and then a solvability condition appears. David Hilbert, who heard Fredholm's results in 1901, spent the next decade on symmetric kernels, and the function spaces he built to do it are what we now call Hilbert spaces.
Integral equations are the natural partner of the differential equations of Chapters 1 and 2. An initial value problem $y''+q(x)y=g$, $y(a)=\alpha$, $y'(a)=\beta$, is equivalent to a Volterra equation (integral from $a$ to $x$); a boundary value problem is equivalent, through its Green's function, to a Fredholm equation (integral over the whole interval). The integral form builds in the initial or boundary conditions, is often easier to analyse — successive approximations converge for every value of the parameter in the Volterra case — and makes the eigenvalue problems of Sturm–Liouville theory into questions about a kernel $K(x,t)$. This chapter develops the classical theory: classification and conversion, kernels of finite rank (where everything reduces to a determinant $\det(I-\lambda A)$), Volterra equations and Laplace transforms, characteristic numbers and eigenfunctions, iterated kernels, the Neumann series and the resolvent kernel, and the spectral theory of symmetric kernels culminating in the Hilbert–Schmidt theorem.
- classify a linear integral equation (first or second kind, Fredholm or Volterra, homogeneous or not), and turn a Volterra equation of the first kind into one of the second kind by differentiation;
- convert an initial value problem into an equivalent Volterra equation and a boundary value problem into an equivalent Fredholm equation (using the Green's function), and convert integral equations back into differential equations with the right initial or boundary conditions;
- solve Fredholm equations with separable kernels by reducing them to a linear system, find their characteristic numbers from $\det(I-\lambda A)=0$, and apply the Fredholm alternative with the transposed equation;
- solve Volterra equations of the second kind by successive approximations and, for convolution kernels, by the Laplace transform, and solve Abel's equation;
- compute iterated kernels, the Neumann series and the resolvent kernel $R(x,t;\lambda)$, state when the series converges ($|\lambda|\,M(b-a)\lt1$, $|\lambda|\,\|K\|_2\lt1$) and why a Volterra resolvent is entire in $\lambda$;
- use the spectral theory of symmetric kernels — real characteristic numbers, orthogonal eigenfunctions, the Hilbert–Schmidt expansion, Mercer's theorem and the trace formulas — to solve equations and to evaluate sums such as $\sum 1/\lambda_n^2=\iint K^2$.
How this is examined
Integral equations form a small but very regular part of the CSIR NET Unit 3 paper, and the questions are mostly short computations. In Part B expect to be asked which integral equation is equivalent to a given initial value problem (watch the factor $(x-t)$ and the terms coming from the initial data), which differential equation and which initial or boundary conditions are hidden in a given integral equation, for the characteristic numbers or the eigenfunctions of a separable kernel such as $xt$, $x+t$, $1+3xt$ or $\cos(x+t)$, for the solution of a Volterra equation with a convolution kernel (a Laplace transform question in disguise), for the resolvent kernel of a simple kernel, or for the value of $\lambda$ that makes the Neumann series converge. Part C, where one or more options may be correct, tests the theory: whether a kernel must have characteristic numbers (Volterra kernels never do; some separable kernels have none, or only complex ones), whether the solvability condition at a characteristic number involves the eigenfunctions of $K$ or of its transpose, whether a Volterra equation of the first kind is always solvable, which properties symmetric kernels have (real characteristic numbers, orthogonal eigenfunctions, the Hilbert–Schmidt expansion) and which they do not. The wrong options are almost always one of the counterexamples collected in the ⚠️ boxes of the two sections. GATE MA sets very similar problems.
Sections in this chapter
Fredholm and Volterra Equations; Separable Kernels
Classification of linear integral equations, conversion of initial and boundary value problems to integral equations and back, separable kernels and the reduction to a linear system, the Fredholm alternative, successive approximations for Volterra equations, convolution kernels, the Laplace transform and Abel's equation.
Eigenvalues, Eigenfunctions and the Resolvent Kernel
Characteristic numbers and eigenfunctions, bounds on them, iterated kernels, the Neumann series and the resolvent kernel, entire resolvents of Volterra kernels, and symmetric kernels: real characteristic numbers, orthogonal eigenfunctions, the Hilbert–Schmidt theorem and Mercer's theorem.
From this course you need: the integral form of an initial value problem, Picard's successive approximations and Gronwall's inequality from §1.1 (Volterra equations are their direct generalisation); existence and uniqueness for linear equations from §2.1; Sturm–Liouville problems, eigenfunction orthogonality and the Rayleigh quotient from §2.3; and above all the Green's function of a two-point boundary value problem and the Fredholm alternative for boundary value problems from §2.4, which turn boundary value problems into Fredholm equations. From the CSIR NET Unit 1 course you need uniform convergence and the Weierstrass M-test (for every series in this chapter), the Cauchy–Schwarz inequality and orthonormal sets with Bessel's inequality in inner product spaces, and from linear algebra determinants, rank, the rank–nullity theorem and the fact that $A$ and $A^{T}$ have the same rank. Laplace transforms and the convolution theorem are used briefly, at the level of a BSc course on differential equations. If this is your first visit, the welcome page explains how each section is organised.