Fredholm and Volterra Equations of the First and Second Kind
CSIR NET integral equations: Fredholm vs Volterra, first vs second kind, IVPs and BVPs as integral equations, Green's functions, Abel's equation.
- 22 min
On this page · 14 sections
- Prerequisites
- The Hook: Putting the Initial Conditions Inside the Equation
- The Standard Forms
- From Initial Value Problems to Volterra Equations
- From Boundary Value Problems to Fredholm Equations
- Solving Volterra Equations of the Second Kind
- Equations of the First Kind
- Memory Tricks
- Don’t Confuse
- Common Mistakes
- Practice
- Quick Revision Table
- Key Takeaways
- What’s Next
Prerequisites
- Riemann, Improper and Lebesgue Integration: the Leibniz rule for differentiating \( \int_a^x F(x, t)\,dt \) is used on every line below.
- Existence and Uniqueness for First-Order IVPs: Picard iteration is already a Volterra integral equation in disguise.
- Laplace transforms of \( 1, x^n, e^{ax}, \sin ax, \cos ax \) and the convolution theorem.
- The chapter map is on Linear Integral Equations.
The Hook: Putting the Initial Conditions Inside the Equation
The Standard Forms
A linear integral equation has the unknown \( \varphi \) appearing linearly, at least once under an integral. With a given kernel \( K(x, t) \), a given free term \( f(x) \) and a parameter \( \lambda \):
| Name | Equation |
|---|---|
| Fredholm, first kind | \( \displaystyle \int_a^b K(x, t)\,\varphi(t)\,dt = f(x) \) |
| Fredholm, second kind | \( \displaystyle \varphi(x) = f(x) + \lambda \int_a^b K(x, t)\,\varphi(t)\,dt \) |
| Volterra, first kind | \( \displaystyle \int_a^x K(x, t)\,\varphi(t)\,dt = f(x) \) |
| Volterra, second kind | \( \displaystyle \varphi(x) = f(x) + \lambda \int_a^x K(x, t)\,\varphi(t)\,dt \) |
Two independent questions settle the type:
- Limits. Both limits fixed: Fredholm. Upper limit equal to the variable \( x \): Volterra.
- Where \( \varphi \) appears. Only under the integral: first kind. Also outside the integral: second kind.
The equation is homogeneous if \( f \equiv 0 \). It is singular if the interval is infinite or the kernel is unbounded (for example \( (x - t)^{-1/2} \) in Abel’s equation below). Some texts also name a third kind, \( h(x)\varphi(x) = f(x) + \lambda\int K\varphi \), where \( h \) vanishes somewhere in the interval; it is outside the scope of the official syllabus line.
graph TD
A["Integral equation in phi"] --> B{"Upper limit fixed or x?"}
B -->|"fixed b"| C["Fredholm"]
B -->|"variable x"| D["Volterra"]
C --> E{"phi outside the integral?"}
D --> F{"phi outside the integral?"}
E -->|no| G["Fredholm 1st kind"]
E -->|yes| H["Fredholm 2nd kind"]
F -->|no| I["Volterra 1st kind"]
F -->|yes| J["Volterra 2nd kind"]
H --> K{"f = 0?"}
K -->|yes| L["Homogeneous: look for characteristic numbers"]A Volterra equation is a special Fredholm equation. Set \( \tilde K(x, t) = K(x, t) \) for \( t \le x \) and \( \tilde K(x, t) = 0 \) for \( t > x \). Then \( \int_a^x K\varphi = \int_a^b \tilde K\varphi \). The triangular shape of \( \tilde K \) is what gives Volterra equations their stronger theory (unique solution for every \( \lambda \)).
Kernels you will meet
| Kernel type | Definition | Example | Why it matters |
|---|---|---|---|
| Symmetric | \( K(x, t) = K(t, x) \) (real) | \( x + t \), \( \min(x, t) \) | real characteristic numbers, orthogonal eigenfunctions |
| Hermitian | \( K(x, t) = \overline{K(t, x)} \) | \( e^{i(x - t)} \) | complex analogue of symmetric |
| Separable (degenerate) | \( \sum_{i=1}^n a_i(x) b_i(t) \) | \( xt \), \( \sin(x + t) \) | reduces to linear algebra; see Separable Kernels |
| Difference (convolution) | \( k(x - t) \) | \( e^{x - t} \), \( \cos(x - t) \) | Volterra version solved by Laplace transform |
| Weakly singular | \( \frac{H(x, t)}{(x - t)^\alpha} \), \( 0 < \alpha < 1 \) | \( (x - t)^{-1/2} \) | Abel’s equation |
From Initial Value Problems to Volterra Equations
The key identity is Cauchy’s formula for repeated integration:
$$\int_a^x \int_a^{x_{n-1}} \cdots \int_a^{x_1} \varphi(t)\,dt\,dx_1 \cdots dx_{n-1} = \int_a^x \frac{(x - t)^{n-1}}{(n - 1)!}\,\varphi(t)\,dt.$$For \( n = 2 \): \( \int_a^x \int_a^s \varphi(t)\,dt\,ds = \int_a^x (x - t)\varphi(t)\,dt \). Swap the order of integration to see it.
Method 1: take \( u = y'' \) as the unknown
For \( y'' + p(x)y' + q(x)y = F(x) \), \( y(a) = \alpha \), \( y'(a) = \beta \), set \( u = y'' \). Then
$$y'(x) = \beta + \int_a^x u(t)\,dt, \qquad y(x) = \alpha + \beta(x - a) + \int_a^x (x - t)\,u(t)\,dt.$$Substituting gives a Volterra equation of the second kind for \( u \):
$$u(x) = F(x) - \beta p(x) - q(x)\big[\alpha + \beta(x - a)\big] - \int_a^x \big[p(x) + q(x)(x - t)\big]u(t)\,dt.$$Method 2: keep \( y \) as the unknown
When there is no \( y' \) term, integrate \( y'' = F - qy \) twice:
$$y(x) = \alpha + \beta(x - a) + \int_a^x (x - t)\big[F(t) - q(t)\,y(t)\big]\,dt.$$So \( y'' + \lambda y = 0 \), \( y(0) = 1 \), \( y'(0) = 0 \) becomes \( y(x) = 1 - \lambda\int_0^x (x - t)\,y(t)\,dt \): the hook equation with \( \lambda = 1 \). An \( n \)-th order IVP gives the kernel \( \frac{(x - t)^{n-1}}{(n - 1)!} \) times the coefficient.
Going back: Volterra equation to IVP
Differentiate with the Leibniz rule
$$\frac{d}{dx}\int_a^x K(x, t)\,\varphi(t)\,dt = K(x, x)\,\varphi(x) + \int_a^x \frac{\partial K}{\partial x}(x, t)\,\varphi(t)\,dt,$$until the integral disappears, and read off the initial conditions by putting \( x = a \) in the equation and its derivatives. This works whenever the kernel is a polynomial in \( x - t \) (or more generally when repeated differentiation in \( x \) kills it or reproduces it).
From Boundary Value Problems to Fredholm Equations
Take \( y'' = g(x) \) on \( [0, 1] \) with \( y(0) = y(1) = 0 \). Integrating twice,
$$y(x) = y'(0)\,x + \int_0^x (x - t)\,g(t)\,dt.$$The condition \( y(1) = 0 \) fixes \( y'(0) = -\int_0^1 (1 - t)\,g(t)\,dt \). Combining the two pieces,
$$y(x) = -\int_0^1 G(x, t)\,g(t)\,dt, \qquad G(x, t) = \begin{cases} t(1 - x), & 0 \le t \le x, \\ x(1 - t), & x \le t \le 1, \end{cases}$$that is, \( G(x, t) = \min(x, t) - xt \). Now put \( g = -\lambda y \) (the problem \( y'' + \lambda y = 0 \)):
$$\boxed{y(x) = \lambda \int_0^1 G(x, t)\,y(t)\,dt.}$$A homogeneous Fredholm equation of the second kind with a symmetric kernel. Its characteristic numbers are the BVP’s eigenvalues \( \lambda_n = n^2\pi^2 \), with eigenfunctions \( \sin n\pi x \). With a forcing term, \( y'' + \lambda y = F \) gives \( y = \lambda\int_0^1 G y\,dt - \int_0^1 G F\,dt \).
\( G \) is the Green’s function of \( -\frac{d^2}{dx^2} \) with those boundary conditions: it is continuous, satisfies the boundary conditions in \( x \), solves \( G'' = 0 \) for \( x \ne t \), and \( \partial G / \partial x \) jumps by \( -1 \) as \( x \) crosses \( t \). Three cases cover almost every exam question on \( [0, 1] \):
| Boundary conditions for \( -y'' = \lambda y \) | Kernel \( G(x, t) \) | Characteristic numbers | Eigenfunctions |
|---|---|---|---|
| \( y(0) = 0 \), \( y(1) = 0 \) | \( \min(x, t) - xt \) | \( n^2\pi^2 \) | \( \sin n\pi x \) |
| \( y(0) = 0 \), \( y'(1) = 0 \) | \( \min(x, t) \) | \( \left(\frac{(2n-1)\pi}{2}\right)^2 \) | \( \sin\frac{(2n-1)\pi x}{2} \) |
| \( y'(0) = 0 \), \( y(1) = 0 \) | \( 1 - \max(x, t) \) | \( \left(\frac{(2n-1)\pi}{2}\right)^2 \) | \( \cos\frac{(2n-1)\pi x}{2} \) |
Solving Volterra Equations of the Second Kind
The proof is successive approximation: the \( n \)-th iterate of the kernel is bounded by \( M^n \frac{(x - t)^{n-1}}{(n-1)!} \), and the factorial beats every power of \( \lambda \). The details, and the resolvent kernel this produces, are on Characteristic Numbers, Eigenfunctions and the Resolvent Kernel. There are three practical methods.
| Method | Use when | Steps |
|---|---|---|
| Reduce to an ODE | kernel polynomial in \( x \) (or \( x - t \)), or exponential | differentiate (Leibniz) until the integral goes; initial values from \( x = a \) |
| Laplace transform | difference kernel \( k(x - t) \), lower limit 0 | \( \Phi(s) = \dfrac{F(s)}{1 - \lambda\,\hat k(s)} \), invert |
| Successive approximation / resolvent | anything, especially to get a series | \( \varphi_{n+1} = f + \lambda\int_a^x K\varphi_n \), or \( \varphi = f + \lambda\int \Gamma f \) |
Equations of the First Kind
Volterra, first kind: differentiate once
For \( \int_a^x K(x, t)\varphi(t)\,dt = f(x) \), the Leibniz rule gives
$$K(x, x)\,\varphi(x) + \int_a^x \frac{\partial K}{\partial x}(x, t)\,\varphi(t)\,dt = f'(x).$$If \( K(x, x) \ne 0 \) on \( [a, b] \), divide by it: a Volterra equation of the second kind, uniquely solvable. Two conditions are built in:
- \( f(a) = 0 \) is necessary (put \( x = a \) in the original equation).
- If \( K(x, x) \equiv 0 \), differentiate again. For \( \int_0^x (x - t)\varphi(t)\,dt = f(x) \) you get \( \varphi = f'' \), and need \( f(0) = f'(0) = 0 \).
Abel’s equation
The classical singular Volterra equation of the first kind is
$$\int_0^x \frac{\varphi(t)}{(x - t)^\alpha}\,dt = f(x), \quad 0 < \alpha < 1, \qquad \text{with solution} \qquad \varphi(x) = \frac{\sin \alpha\pi}{\pi}\,\frac{d}{dx}\int_0^x \frac{f(t)}{(x - t)^{1 - \alpha}}\,dt.$$For \( \alpha = \frac{1}{2} \) the factor is \( \frac{1}{\pi} \). The fastest route in an exam is Laplace: \( \mathcal{L}\{x^{-\alpha}\} = \Gamma(1 - \alpha)\,s^{\alpha - 1} \), so \( \Phi(s) = \frac{F(s)\,s^{1 - \alpha}}{\Gamma(1 - \alpha)} \). Or guess a power: \( \int_0^x \frac{t^\beta}{(x - t)^{1/2}}\,dt = B\!\left(\beta + 1, \tfrac{1}{2}\right)x^{\beta + 1/2} \).
Fredholm, first kind: the badly behaved one
\( \int_a^b K(x, t)\varphi(t)\,dt = f(x) \) has no general existence or uniqueness theory comparable to the second kind:
- Existence can fail. With \( K = xt \) on \( [0, 1] \), the left side is always a multiple of \( x \). So \( \int_0^1 xt\,\varphi(t)\,dt = x^2 \) has no solution at all.
- Uniqueness can fail. \( \int_0^1 xt\,\varphi(t)\,dt = x \) needs only \( \int_0^1 t\varphi(t)\,dt = 1 \): \( \varphi = 2 \) works, and so does \( 2 + \psi \) for any \( \psi \) with \( \int_0^1 t\psi = 0 \).
- Instability. For a continuous kernel, \( \int_a^b K(x, t)\sin(nt)\,dt \to 0 \) as \( n \to \infty \) (Riemann–Lebesgue). A large, rapidly oscillating change in \( \varphi \) produces a tiny change in \( f \), so small errors in \( f \) can mean huge errors in \( \varphi \).
Part C likes exactly these three points. For a separable kernel the rule is simple: a first-kind equation is solvable only if \( f \) lies in the span of the \( a_i(x) \).
Memory Tricks
Don’t Confuse
| This | vs | That |
|---|---|---|
| Volterra, 2nd kind: unique solution for all \( \lambda \) | Fredholm, 2nd kind: unique only if \( \lambda \) is not a characteristic number | |
| Volterra, 1st kind: differentiate to get 2nd kind (needs \( K(x, x) \ne 0 \), \( f(a) = 0 \)) | Fredholm, 1st kind: may have no solution or infinitely many | |
| \( \int_0^x (x - t)\varphi(t)\,dt \) (double integral) | \( x\int_0^x \varphi(t)\,dt \) (not the same) | |
| Leibniz rule term \( K(x, x)\varphi(x) \) | forgetting it (the most common slip) | |
| \( \min(x, t) - xt \): \( y(0) = y(1) = 0 \) | \( \min(x, t) \): \( y(0) = 0 \), \( y'(1) = 0 \) |
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
- Classify \( \varphi(x) = x + \int_0^1 e^{xt}\varphi(t)\,dt \).
- Classify \( \int_0^x \cos(x - t)\,\varphi(t)\,dt = x \).
- Write \( \int_0^x \int_0^s \varphi(t)\,dt\,ds \) as a single integral.
Answers
- Fredholm, second kind, non-homogeneous, \( \lambda = 1 \), kernel \( e^{xt} \) (symmetric, not separable).
- Volterra, first kind, difference kernel. \( f(0) = 0 \) and \( K(x, x) = 1 \ne 0 \), so differentiating gives a second-kind equation.
- \( \int_0^x (x - t)\,\varphi(t)\,dt \).
Level 2: Part B Style
Practice Q1. The continuous solution of \( \varphi(x) = 1 - \int_0^x (x - t)\,\varphi(t)\,dt \) is:
(a) \( \sin x \) (b) \( \cos x \) (c) \( e^{-x} \) (d) \( \cosh x \)
Answer: (b)
\( \varphi(0) = 1 \) rules out (a). Differentiating twice: \( \varphi' = -\int_0^x \varphi \), \( \varphi'' = -\varphi \), with \( \varphi(0) = 1 \), \( \varphi'(0) = 0 \). So \( \varphi = \cos x \). (c) has \( \varphi'(0) = -1 \); (d) solves \( \varphi'' = +\varphi \).
Practice Q2. The initial value problem \( y'' + y = 0 \), \( y(0) = 0 \), \( y'(0) = 1 \) is equivalent to:
(a) \( y(x) = x - \int_0^x (x - t)\,y(t)\,dt \) (b) \( y(x) = 1 - \int_0^x (x - t)\,y(t)\,dt \) (c) \( y(x) = x + \int_0^x (x - t)\,y(t)\,dt \) (d) \( y(x) = x - \int_0^1 (x - t)\,y(t)\,dt \)
Answer: (a)
\( y'' = -y \) integrated twice with \( y(0) = 0 \), \( y'(0) = 1 \) gives \( y = 0 + 1 \cdot x - \int_0^x (x - t)y(t)\,dt \). (b) has the wrong initial data (\( y(0) = 1 \), \( y'(0) = 0 \)); (c) is \( y'' = +y \); (d) is not a Volterra equation and does not encode the IVP.
Practice Q3. The continuous solution of \( \int_0^x e^{x - t}\,\varphi(t)\,dt = x^2 \) is:
(a) \( 2x \) (b) \( 2x - x^2 \) (c) \( x^2 - 2x \) (d) \( 2 - 2x \)
Answer: (b)
Differentiate: \( \varphi(x) + \int_0^x e^{x - t}\varphi(t)\,dt = 2x \). The integral is \( x^2 \), so \( \varphi = 2x - x^2 \). Check: \( \int_0^x e^{x - t}(2t - t^2)\,dt = x^2 \) (verified by direct integration).
Practice Q4. The boundary value problem \( y'' + \lambda y = 0 \), \( y(0) = 0 \), \( y'(1) = 0 \) is equivalent to \( y(x) = \lambda\int_0^1 K(x, t)\,y(t)\,dt \) with \( K(x, t) = \)
(a) \( \min(x, t) - xt \) (b) \( \min(x, t) \) (c) \( 1 - \max(x, t) \) (d) \( \max(x, t) \)
Answer: (b)
\( G(x, t) = \min(x, t) \) is \( x \) for \( x < t \) and \( t \) for \( x > t \). It vanishes at \( x = 0 \), is constant (so \( \partial_x G = 0 \)) for \( x > t \), which gives \( y'(1) = 0 \), and its slope drops by 1 at \( x = t \). (a) is for \( y(0) = y(1) = 0 \); (c) is for \( y'(0) = 0 \), \( y(1) = 0 \); (d) does not vanish at \( x = 0 \).
Level 3: Part C Style
Practice Q5. Consider \( \varphi(x) = \sin x + 2\int_0^x \cos(x - t)\,\varphi(t)\,dt \). Which statements are true?
(a) It is a Volterra equation of the second kind with a difference kernel. (b) Its continuous solution is \( \varphi(x) = x e^x \). (c) The homogeneous equation \( \psi(x) = 2\int_0^x \cos(x - t)\psi(t)\,dt \) has a non-zero continuous solution. (d) \( \varphi'(0) = 1 \).
Answer: (a), (b) and (d)
- (a) True. Variable upper limit, \( \varphi \) outside the integral, kernel depends on \( x - t \) only.
- (b) True. Laplace: \( \Phi = \frac{1}{s^2 + 1} + \frac{2s}{s^2 + 1}\Phi \) gives \( \Phi = \frac{1}{(s - 1)^2} \).
- (c) False. A Volterra equation with continuous kernel has only the zero solution in the homogeneous case, for every \( \lambda \).
- (d) True. \( \varphi'(x) = (1 + x)e^x \), so \( \varphi'(0) = 1 \). From the equation directly: \( \varphi'(x) = \cos x + 2\varphi(x) - 2\int_0^x \sin(x - t)\varphi \), so \( \varphi'(0) = 1 + 0 - 0 = 1 \).
Practice Q6. Which statements about equations of the first kind are true? (All functions are continuous on \( [0, 1] \).)
(a) If \( \int_0^x \varphi(t)\,dt = f(x) \) has a continuous solution, then \( f(0) = 0 \). (b) \( \int_0^x \frac{\varphi(t)}{\sqrt{x - t}}\,dt = \pi x \) has the solution \( \varphi(x) = 2\sqrt{x} \). (c) \( \int_0^1 xt\,\varphi(t)\,dt = x^2 \) has a continuous solution. (d) \( \int_0^1 xt\,\varphi(t)\,dt = x \) has infinitely many continuous solutions.
Answer: (a), (b) and (d)
- (a) True. Put \( x = 0 \).
- (b) True. \( \int_0^x \frac{2\sqrt t}{\sqrt{x - t}}\,dt = 2B\!\left(\tfrac{3}{2}, \tfrac{1}{2}\right)x = 2 \cdot \frac{\pi}{2}x = \pi x \).
- (c) False. The left side is \( x \cdot \int_0^1 t\varphi(t)\,dt \), a multiple of \( x \); it can never equal \( x^2 \).
- (d) True. Any \( \varphi \) with \( \int_0^1 t\varphi(t)\,dt = 1 \) works: \( \varphi = 2 \), \( \varphi = 2 + c(3t - 2) \) for every \( c \) (since \( \int_0^1 t(3t - 2)\,dt = 0 \)).
Practice Q7. The problem \( y'' + \lambda y = 0 \), \( y(0) = y(1) = 0 \) is written as \( y(x) = \lambda\int_0^1 G(x, t)\,y(t)\,dt \). Which statements are true?
(a) \( G(x, t) = \min(x, t) - xt \). (b) \( G(x, t) = G(t, x) \) for all \( x, t \in [0, 1] \). (c) The characteristic numbers of the integral equation are \( \lambda = n\pi \), \( n = 1, 2, \dots \) (d) \( \int_0^1 G(x, x)\,dx = \frac{1}{6} \).
Answer: (a), (b) and (d)
- (a) True. Derived on this page: \( t(1 - x) \) for \( t \le x \), \( x(1 - t) \) for \( t \ge x \).
- (b) True. \( \min(x, t) - xt \) is visibly symmetric.
- (c) False. The non-trivial solutions are \( \sin n\pi x \) with \( \lambda = n^2\pi^2 \), not \( n\pi \).
- (d) True. \( G(x, x) = x - x^2 \), and \( \int_0^1 (x - x^2)\,dx = \frac{1}{2} - \frac{1}{3} = \frac{1}{6} \). This equals \( \sum \frac{1}{n^2\pi^2} = \frac{1}{6} \), the trace formula on the resolvent page.
Quick Revision Table
| Idea | Fact |
|---|---|
| Fredholm vs Volterra | fixed limits vs upper limit \( x \) |
| First vs second kind | \( \varphi \) only inside vs also outside |
| Volterra as Fredholm | kernel set to 0 for \( t > x \) |
| Repeated integration | \( n \)-fold \( = \int_a^x \frac{(x - t)^{n-1}}{(n-1)!}\varphi \) |
| IVP \( y'' = F - qy \) | \( y = \alpha + \beta x + \int_0^x (x - t)[F - qy] \) |
| BVP \( y(0) = y(1) = 0 \) | kernel \( \min(x, t) - xt \), \( \lambda_n = n^2\pi^2 \) |
| BVP \( y(0) = 0 = y'(1) \) | kernel \( \min(x, t) \), \( \lambda_n = \left(\frac{(2n-1)\pi}{2}\right)^2 \) |
| Leibniz rule | \( \frac{d}{dx}\int_a^x K\varphi = K(x, x)\varphi(x) + \int_a^x K_x\varphi \) |
| Volterra 2nd kind | unique solution for every \( \lambda \); no characteristic numbers |
| Laplace (difference kernel) | \( \Phi = F/(1 - \lambda\hat k) \) |
| Volterra 1st kind | needs \( f(a) = 0 \); differentiate; needs \( K(x, x) \ne 0 \) |
| Abel, \( \alpha = \frac{1}{2} \) | \( \varphi = \frac{1}{\pi}\frac{d}{dx}\int_0^x \frac{f(t)}{\sqrt{x - t}}\,dt \) |
| Fredholm 1st kind | may have no solution or infinitely many; ill-posed |
Key Takeaways
What’s Next
- Separable Kernels and the Fredholm Alternative: Fredholm equations of the second kind with degenerate kernels, solved as linear systems.
- Characteristic Numbers, Eigenfunctions and the Resolvent Kernel: iterated kernels, the Neumann series and symmetric kernels.
- Linear Integral Equations: Quick Revision.
- The Picard iteration on Existence and Uniqueness for First-Order IVPs is the same successive approximation applied to a non-linear Volterra equation.