Separable Kernels and the Fredholm Alternative
CSIR NET integral equations: solving Fredholm equations with separable (degenerate) kernels as linear systems, characteristic numbers, Fredholm alternative.
- 20 min
On this page · 13 sections
Prerequisites
- Fredholm and Volterra Equations of the First and Second Kind: the standard forms and vocabulary.
- Rank, Determinants and Linear Equations: a separable kernel turns the integral equation into \( (I - \lambda A)c = F \), and solvability is a rank question.
- Eigenvalues and the Cayley–Hamilton Theorem: characteristic numbers are reciprocals of the non-zero eigenvalues of the matrix \( A \).
The Hook: An Integral Equation That Is Really a 2 by 2 System
Separable Kernels
A kernel is separable (or degenerate) if it is a finite sum of products
$$K(x, t) = \sum_{i=1}^{n} a_i(x)\,b_i(t),$$where we may assume \( a_1, \dots, a_n \) are linearly independent, and so are \( b_1, \dots, b_n \). Then \( n \) is the rank of the kernel: the integral operator maps every function into the \( n \)-dimensional span of the \( a_i \).
| Kernel on the given interval | Separable? | Splitting |
|---|---|---|
| \( xt \) | yes, rank 1 | \( a = x \), \( b = t \) |
| \( e^{x + t} \) | yes, rank 1 | \( e^x \cdot e^t \) |
| \( x + t \) | yes, rank 2 | \( x \cdot 1 + 1 \cdot t \) |
| \( \sin(x + t) \) | yes, rank 2 | \( \sin x\cos t + \cos x\sin t \) |
| \( \cos(x - t) \) | yes, rank 2 | \( \cos x\cos t + \sin x\sin t \) |
| \( 1 + 3xt \) | yes, rank 2 | \( 1 \cdot 1 + x \cdot 3t \) |
| \( (1 + xt)^2 \) | yes, rank 3 | \( 1 + 2xt + x^2t^2 \) |
| \( e^{xt} \) | no | \( \sum \frac{x^k t^k}{k!} \) needs infinitely many terms |
| \( \min(x, t) \), \( \lvert x - t \rvert \) | no | infinitely many characteristic numbers (see below) |
The Method
For \( \varphi(x) = f(x) + \lambda\int_a^b K(x, t)\varphi(t)\,dt \) with \( K = \sum a_i(x)b_i(t) \):
- Name the unknown constants. \( c_i = \int_a^b b_i(t)\,\varphi(t)\,dt \). Then $$\varphi(x) = f(x) + \lambda\sum_{j=1}^{n} c_j\,a_j(x).$$
- Substitute back. Multiply by \( b_i(x) \) and integrate: $$c_i = F_i + \lambda\sum_{j=1}^{n} A_{ij}\,c_j, \qquad A_{ij} = \int_a^b b_i(t)\,a_j(t)\,dt, \qquad F_i = \int_a^b b_i(t)\,f(t)\,dt.$$
- Solve the linear system \( (I - \lambda A)\,c = F \).
- Assemble \( \varphi = f + \lambda\sum c_j a_j \).
The characteristic numbers are the roots of
$$\boxed{D(\lambda) = \det(I - \lambda A) = 0.}$$\( D(\lambda) \) is a polynomial of degree at most \( n \) with \( D(0) = 1 \). Its roots are \( \lambda = \frac{1}{\mu} \) for the non-zero eigenvalues \( \mu \) of \( A \). An eigenvector \( c \) of \( A \) for \( \mu \) gives the eigenfunction \( \varphi = \sum c_j a_j \).
graph TD
A["phi = f + lambda * integral of K phi, K = sum a_i(x) b_i(t)"] --> B["c_i = integral of b_i phi; phi = f + lambda * sum c_j a_j"]
B --> C["Build A_ij = integral of b_i a_j and F_i = integral of b_i f"]
C --> D{"det(I - lambda A) = 0 ?"}
D -->|"no"| E["Unique solution for every f: solve (I - lambda A)c = F"]
D -->|"yes: lambda is a characteristic number"| F{"F orthogonal to solutions of transposed system?"}
F -->|"yes"| G["Infinitely many solutions: particular + any eigenfunction"]
F -->|"no"| H["No solution"]Worked examples
The Fredholm Alternative
For a separable kernel, everything follows from the linear algebra of \( (I - \lambda A)c = F \). The general theorem (true for any continuous or square-integrable kernel, as Fredholm proved) has the same shape.
For a complex kernel the “transposed” equation uses the adjoint kernel \( \overline{K(t, x)} \) and the condition \( \int f\overline{\psi} = 0 \). For a symmetric kernel, the transposed equation is the original one, so the condition reads: \( f \) is orthogonal to every eigenfunction for that \( \lambda \).
Multiplicity
The multiplicity (or index) of a characteristic number is the number of linearly independent eigenfunctions. For a separable kernel it is \( n - \operatorname{rank}(I - \lambda A) \), the geometric multiplicity of the eigenvalue \( \frac{1}{\lambda} \) of \( A \). In Worked example 4, \( \cos(x - t) \) has one characteristic number of multiplicity 2.
First-Kind Equations with Separable Kernels
For \( \int_a^b \sum a_i(x)b_i(t)\varphi(t)\,dt = f(x) \), the left side is \( \sum c_i a_i(x) \). So:
- Necessary: \( f \) lies in the span of \( a_1, \dots, a_n \).
- If it does, \( f = \sum \gamma_i a_i \), any \( \varphi \) with \( \int b_i\varphi = \gamma_i \) for all \( i \) works, and there are infinitely many (add anything orthogonal to all the \( b_i \)).
This is the cleanest illustration of why Fredholm equations of the first kind are ill-posed.
Memory Tricks
Don’t Confuse
| This | vs | That |
|---|---|---|
| characteristic number \( \lambda \) (homogeneous \( \varphi = \lambda\int K\varphi \) solvable) | eigenvalue \( \mu = \frac{1}{\lambda} \) of the operator or of \( A \) | |
| eigenfunction of \( K(x, t) \) | eigenfunction of the transposed kernel \( K(t, x) \) (used in the solvability test) | |
| \( \det(I - \lambda A) \) | \( \det(A - \lambda I) \) (gives eigenvalues of \( A \), not characteristic numbers) | |
| no characteristic numbers (\( \sin x\cos t \) on \( [0, \pi] \)) | \( \lambda = 0 \) (never a characteristic number: \( \varphi = 0 \cdot \int K\varphi \) forces \( \varphi = 0 \)) | |
| rank of the kernel \( n \) | number of characteristic numbers (\( \le n \), counted without multiplicity) |
Common Mistakes
Practice
All questions below are original practice questions written for this page. They are not from any CSIR NET paper. In Level 3, as in Part C of Mathematical Sciences, one or more options may be correct, and each option is justified in the answer.
Level 1: Foundation
- Is \( e^{xt} \) a separable kernel on \( [0, 1] \)?
- Find the characteristic number of \( \varphi(x) = \lambda\int_0^1 e^{x + t}\varphi(t)\,dt \).
- At most how many characteristic numbers can \( K(x, t) = x^2t + xt^2 \) have?
Answers
- No. Its expansion \( \sum_k \frac{x^k t^k}{k!} \) has infinitely many independent terms; it is symmetric and has infinitely many characteristic numbers.
- \( A = \left[\int_0^1 e^{2t}\,dt\right] = \left[\frac{e^2 - 1}{2}\right] \), so \( \lambda = \frac{2}{e^2 - 1} \approx 0.313 \), eigenfunction \( e^x \).
- Two (rank 2). Here \( A = \begin{pmatrix} \frac{1}{4} & \frac{1}{3} \\ \frac{1}{5} & \frac{1}{4} \end{pmatrix} \) with \( a = (x^2, x) \), \( b = (t, t^2) \), and \( \det(I - \lambda A) = 0 \) gives \( \lambda^2 + 120\lambda - 240 = 0 \), so \( \lambda = -60 \pm 16\sqrt{15} \): exactly two.
Level 2: Part B Style
Practice Q1. For \( \lambda \ne 3 \), the solution of \( \varphi(x) = x + \lambda\int_0^1 xt\,\varphi(t)\,dt \) is:
(a) \( \frac{3x}{3 - \lambda} \) (b) \( \frac{x}{1 - 3\lambda} \) (c) \( x + \frac{\lambda}{3} \) (d) \( \frac{3x}{3 + \lambda} \)
Answer: (a)
Worked example 1: \( c = \int_0^1 t\varphi = \frac{1}{3 - \lambda} \), \( \varphi = x(1 + \lambda c) = \frac{3x}{3 - \lambda} \). Quick check: it must blow up at the characteristic number \( \lambda = 3 \), which only (a) does.
Practice Q2. The characteristic numbers of \( K(x, t) = \sin(x + t) \) on \( [0, 2\pi] \) are:
(a) \( \pm\pi \) (b) \( \pm\frac{1}{\pi} \) (c) \( \frac{1}{\pi} \) only (d) none
Answer: (b)
\( A = \begin{pmatrix} 0 & \pi \\ \pi & 0 \end{pmatrix} \), \( \det(I - \lambda A) = 1 - \lambda^2\pi^2 \). Roots \( \lambda = \pm\frac{1}{\pi} \). Option (a) is the eigenvalues of \( A \), not the characteristic numbers.
Practice Q3. The number of characteristic numbers of \( K(x, t) = \sin x\cos t \) on \( [0, \pi] \) is:
(a) 0 (b) 1 (c) 2 (d) infinitely many
Answer: (a)
Rank 1 with \( \int_0^\pi \cos t\sin t\,dt = 0 \), so \( \det(I - \lambda A) \equiv 1 \). (On \( [0, \frac{\pi}{2}] \) the integral is \( \frac{1}{2} \) and there would be exactly one, \( \lambda = 2 \): always recompute on the given interval.)
Practice Q4. The equation \( \varphi(x) = f(x) + 2\int_0^1 x\,\varphi(t)\,dt \), with \( f \) continuous, has a solution if and only if:
(a) \( \int_0^1 f(x)\,dx = 0 \) (b) \( \int_0^1 x f(x)\,dx = 0 \) (c) \( f(0) = 0 \) (d) always
Answer: (a)
\( \lambda = 2 \) is the characteristic number (Worked example 6). The transposed kernel \( K(t, x) = t \) gives constant \( \psi \), so the condition is \( f \perp 1 \). Option (b) is the trap of testing against the eigenfunction \( x \) of the original kernel.
Level 3: Part C Style
Practice Q5. Consider the homogeneous equation \( \varphi(x) = \lambda\int_0^1 (x + t)\,\varphi(t)\,dt \). Which statements are true?
(a) The characteristic numbers are \( -6 \pm 4\sqrt{3} \). (b) Both characteristic numbers are real. (c) Eigenfunctions belonging to the two characteristic numbers are orthogonal on \( [0, 1] \). (d) The product of the characteristic numbers is 12.
Answer: (a), (b) and (c)
- (a) True. \( \det(I - \lambda A) = 1 - \lambda - \frac{\lambda^2}{12} \), roots of \( \lambda^2 + 12\lambda - 12 = 0 \).
- (b) True. Discriminant \( 144 + 48 > 0 \). (Real is guaranteed anyway: the kernel is real and symmetric.)
- (c) True. \( (\sqrt3 x + 1) \) and \( (\sqrt3 x - 1) \): \( \int_0^1 (3x^2 - 1)\,dx = 0 \).
- (d) False. The product of the roots of \( \lambda^2 + 12\lambda - 12 \) is \( -12 \). One characteristic number is positive, one negative.
Practice Q6. Consider \( \varphi(x) = f(x) + \lambda\int_{-1}^{1} (1 + 3xt)\,\varphi(t)\,dt \). Which statements are true?
(a) \( \lambda = \frac{1}{2} \) is the only characteristic number. (b) For \( \lambda = \frac{1}{2} \) the homogeneous equation has two linearly independent solutions. (c) For \( \lambda = \frac{1}{2} \) and \( f(x) = x^2 \), the equation has a solution. (d) For \( \lambda = 1 \) and \( f(x) = 1 \), the solution is \( \varphi(x) = -1 \).
Answer: (a), (b) and (d)
With \( a = (1, x) \), \( b = (1, 3t) \): \( A = \begin{pmatrix} \int_{-1}^1 1 & \int_{-1}^1 t \\ \int_{-1}^1 3t & \int_{-1}^1 3t^2 \end{pmatrix} = \begin{pmatrix} 2 & 0 \\ 0 & 2 \end{pmatrix} \).
- (a) True. \( \det(I - \lambda A) = (1 - 2\lambda)^2 \).
- (b) True. \( I - \frac{1}{2}A = 0 \), rank 0, so the eigenspace is all of \( \operatorname{span}\{1, x\} \).
- (c) False. The kernel is symmetric, so \( f \) must be orthogonal to \( 1 \) and \( x \) on \( [-1, 1] \). \( \int_{-1}^1 x^2\,dx = \frac{2}{3} \ne 0 \). (By contrast \( f = x^2 - \frac{1}{3} \) passes both tests and is solvable.)
- (d) True. \( \varphi = 1 + c_1 + c_2 x \) with \( c_1 = \int\varphi = 2 + 2c_1 \Rightarrow c_1 = -2 \), and \( c_2 = \int 3t\varphi = 2c_2 \Rightarrow c_2 = 0 \). So \( \varphi = -1 \).
Practice Q7. Consider \( \varphi(x) = f(x) + \lambda\int_0^{2\pi} \cos(x - t)\,\varphi(t)\,dt \). Which statements are true?
(a) \( \lambda = \frac{1}{\pi} \) is the only characteristic number. (b) The characteristic number \( \frac{1}{\pi} \) has multiplicity 2. (c) For \( \lambda \ne \frac{1}{\pi} \) and \( f(x) = \cos x \), the solution is \( \varphi(x) = \frac{\cos x}{1 - \lambda\pi} \). (d) For \( \lambda = \frac{1}{\pi} \) and \( f(x) = 1 + \sin x \), the equation has a solution.
Answer: (a), (b) and (c)
- (a) True and (b) True. \( A = \pi I \) (Worked example 4), so \( \det(I - \lambda A) = (1 - \lambda\pi)^2 \), and at \( \lambda = \frac{1}{\pi} \) both \( \cos x \) and \( \sin x \) are eigenfunctions.
- (c) True. \( c_1 = \int\cos t\,\varphi = \pi + \lambda\pi c_1 \), \( c_2 = \int\sin t\,\varphi = \lambda\pi c_2 = 0 \). So \( \varphi = \cos x + \frac{\lambda\pi\cos x}{1 - \lambda\pi} = \frac{\cos x}{1 - \lambda\pi} \).
- (d) False. The kernel is symmetric, so \( f \) must be orthogonal to \( \cos x \) and \( \sin x \). \( \int_0^{2\pi} (1 + \sin x)\sin x\,dx = \pi \ne 0 \). (With \( f = 1 \) alone, it would be solvable: \( \varphi = 1 + C_1\cos x + C_2\sin x \).)
Quick Revision Table
| Idea | Fact |
|---|---|
| Separable kernel | \( K = \sum_{i=1}^n a_i(x)b_i(t) \), rank \( n \) |
| Unknowns | \( c_i = \int b_i\varphi \); \( \varphi = f + \lambda\sum c_j a_j \) |
| Matrix | \( A_{ij} = \int b_i a_j \), \( F_i = \int b_i f \), \( (I - \lambda A)c = F \) |
| Characteristic numbers | roots of \( \det(I - \lambda A) = 0 \); \( \lambda = 1/\mu \), \( \mu \ne 0 \) eigenvalue of \( A \) |
| How many | at most \( n \); possibly none |
| Rank 1, \( a(x)b(t) \) | \( \lambda = 1/\int ab \) if \( \int ab \ne 0 \); else none |
| \( xt \) on \( [0, 1] \) | \( \lambda = 3 \), eigenfunction \( x \) |
| \( x + t \) on \( [0, 1] \) | \( \lambda = -6 \pm 4\sqrt3 \), eigenfunctions \( \sqrt3 x \pm 1 \) |
| \( \sin(x + t) \), \( [0, 2\pi] \) | \( \lambda = \pm\frac{1}{\pi} \) |
| \( \cos(x - t) \), \( [0, 2\pi] \) | \( \lambda = \frac{1}{\pi} \), multiplicity 2 |
| Fredholm alternative | not characteristic: unique solution for all \( f \); characteristic: solvable iff \( f \perp \) transposed eigenfunctions, then infinitely many |
| First kind, separable | solvable only if \( f \in \operatorname{span}\{a_i\} \); never unique |
Key Takeaways
What’s Next
- Characteristic Numbers, Eigenfunctions and the Resolvent Kernel: the resolvent for separable kernels in closed form, the Neumann series, and the theory of symmetric kernels.
- Inner Product Spaces and Orthonormal Bases: the orthogonality language used in the Fredholm alternative and the spectral theorem behind symmetric kernels.
- Linear Integral Equations: Quick Revision.