An integral equation asks for a function that appears inside an integral. Writing it as turns the problem into invertibility of an operator, but the shape of the integration region matters. A norm-small operator gives a geometric-series solution. For Volterra equations, ordered integration yields a stronger factorial bound, so the series converges even when the first operator norm is large. A Fredholm equation can instead have a nontrivial kernel at a particular parameter; then compatibility replaces unconditional solvability.
We work over the real or complex scalar field on with the supremum norm, where . This is a Banach space. All kernels and input functions below are continuous on the stated compact domains; integrals are ordinary Riemann integrals, also equal to their Lebesgue counterparts.
Bounded integral operators and Neumann series
Definition 2.4.1
(Fredholm integral operator with continuous kernel)
Let be continuous on the square . The Fredholm integral operator with kernel is
For a scalar and , the equation is a Fredholm equation of the second kind.
Uniform continuity of on the square shows is continuous. If and , then , so is bounded. The estimate is generally not its exact norm.
Theorem 2.4.2
(Neumann series)
Let be a bounded linear operator on a Banach space with . Then is invertible and
Proof
(Geometric partial sums in a complete operator space)
The partial sums are Cauchy in operator norm because and ; the space of bounded operators on a Banach space is complete, so . Multiplying the finite sum on either side by gives , and . Passing to the limit gives , a two-sided inverse. Finally .
Thus has a unique solution when . The computable condition is sufficient, though possibly stronger than needed. A norm bound below one is not a necessary condition for invertibility.
The factorial advantage of a Volterra operator
Definition 2.4.3
(Volterra integral operator)
Let be continuous on the triangle . The Volterra operator with kernel is
The upper limit of integration is the variable itself: the value at time uses only on .
The output is continuous: when changes, uniform continuity controls the integrand on the shared part of the integration interval, while boundedness controls the short added or removed part.
Lemma 2.4.4
(Factorial bound for Volterra powers)
For every , every and every ,
In particular with .
Proof
(Induction over the ordered integration region)
For , . If the bound holds for , then
Taking the supremum over gives the operator-norm bound.
The factorial comes from the ordered integration region, not from assuming that .
Theorem 2.4.5
(Unique solvability of Volterra equations of the second kind)
For every scalar and every , the equation
has a unique solution in , with
Proof
(Summing the factorial majorant)
The factorial lemma gives . Thus the operator series converges absolutely in operator norm, and its sum is a bounded operator. Its finite partial sums satisfy , as does multiplication in the reverse order. The remainder tends to zero by the factorial estimate, so . Hence is invertible, the unique solution is , and summing the majorant gives .
Proof X-ray
Why a large first norm does not prevent convergence
The argument has three ingredients. Continuity of on the closed triangle makes finite; completeness of the bounded-operator space turns an absolutely convergent series into an operator; and the telescoping identity on both sides turns that operator into a two-sided inverse.
The one non-obvious move is in the lemma. A geometric estimate would replace by and could fail when . The nested upper limits restrict the -fold integration to the region , whose volume is , so the actual powers obey . Since converges for every , no smallness of is needed. For a full-interval kernel the same induction integrates over the whole cube and only gives , which is why the Fredholm case needs a different analysis.
The first example shows the theorem working where the simple norm test fails.
Worked example 2.4.6
(Constant kernel with )
On , take , , and . Here (the bound is attained by at ), so and the norm-small condition fails. But , and the Volterra theorem gives
Directly, . Failing a sufficient condition does not imply nonexistence or divergence.
Worked example 2.4.7
(The kernel and a hyperbolic cosine)
Solve on . With and , compute the powers on . First . If , then
Hence . Check: , so the right-hand side is . The estimate with , is consistent with .
Compactness of a continuous Fredholm kernel
Proposition 2.4.8
(Continuous kernels give compact operators)
If is continuous on , the Fredholm operator is compact on : it sends the unit ball to a relatively compact set.
Proof
(Equicontinuity from the kernel, then Arzelà–Ascoli)
The images of the unit ball are uniformly bounded by . For ,
Uniform continuity of makes this bound tend to zero as , independently of . The image is equicontinuous and uniformly bounded, so Arzelà–Ascoli gives a uniformly convergent subsequence from every sequence in the image.
Compactness does not make invertible for every ; the next section exhibits the failure exactly.
A finite-rank Fredholm equation
Suppose , where the functions are continuous. Such a kernel is called degenerate, and then has range in .
Proposition 2.4.9
(Reduction of a degenerate kernel to a linear system)
Put and . The formula
gives a bijection between solutions of and solutions of . This holds even if the chosen are linearly dependent.
Proof
(Integrate against each )
If solves the integral equation, put . Then , so has the displayed form; multiplying by and integrating gives , that is, .
Conversely, let solve the finite system and define by the formula. Then for every , so and . Since is recovered from as , distinct solutions give distinct functions .
Thus finite-dimensional linear algebra gives both existence and all solutions.
Common mistake
(Transposing the coefficient matrix)
The entry pairs the test function with the range function : . Writing instead produces the transpose. Its determinant is the same, so the resonant values of do not change, but the system has different solutions in general, and the reconstructed is then wrong.
Worked example 2.4.10
(Rank-one resonance)
On , let (so , , ) and set . Writing gives . If , the unique solution is . At , a solution exists exactly when , and then every solution is with arbitrary scalar . For there is no solution at resonance; for , there are infinitely many. These are exact compatibility statements, not deductions from a failed norm estimate.
Common mistake
(Using the generic formula at the resonant parameter)
The formula is derived by dividing by , so it says nothing at . There one must return to , which reads : either no solution exists, or the solution is determined only up to an added constant. Reporting a single solution at resonance loses the whole kernel of .
The same rank-one operator also shows that invertibility can hold far outside the norm-small regime.
Counterexample mode
A norm test is sufficient, not necessary
Claim to test: if , then is not invertible. For this rank-one operator , : the integral estimate gives , and attains equality. Also . At , gives , so the equation has the unique solution even though . The Neumann series does not converge in operator norm: applied to the constant one, its terms for are . The correct statement is one-directional: guarantees invertibility, while outside that regime invertibility must be decided by other means, here by the scalar equation.
The general Fredholm alternative for compact operators extends the resonance picture beyond finite rank, but it is not proved here. Compactness alone is not a license to assume an explicit finite matrix, nor does the Volterra factorial estimate apply to arbitrary full-interval kernels.
Concept lensAnalytic
Solvability read from a power series in the parameter
Both cases describe the operator-valued function , with or . For a Volterra operator, , so has infinite radius of convergence: one power series gives the inverse for every scalar . For the rank-one operator with , direct multiplication gives
a rational function whose only singularity is the resonant value . Its power series converges in norm only for , which is the counterexample above seen from this side.
Practice with solutions
Quick check
(On , solve using the Volterra series.)
Compute for the constant kernel.
Solution · Answer
With , , so . Direct integration verifies it: .
Quick check
(For , solve .)
Reduce to one scalar equation for .
Solution · Answer
Here and , so and . Check: , and .
Quick check
(For the same at , does have a solution? Does have a unique solution?)
Compute the compatibility number in each case.
Solution · Answer
For , , so the equation has no solution. For , , so every with arbitrary scalar is a solution; the solution exists but is not unique.
Quick check
(If on the Volterra triangle over , bound and explain why the series converges when .)
Apply the factorial lemma with , .
Solution · Answer
The lemma gives . Multiplying by yields the majorant , whose sum is , so the series converges absolutely even though is far above one.