I would like to compute the eigenvalues and eigenfunctions for the Fredholm integral operator

$$K(g) = \int_0^1 e^{xt} g(t) \,dt.$$

The sources I've checked* seem to say that the process is fairly involved. Has anything been published on this kernel? Or, if not, am I correct that it's going to be a hard thing to do?

* See, e.g., equations (12) and on here: https://www.encyclopediaofmath.org/index.php/Fredholm_equation

  • $\begingroup$ $k$ is operating on which space? $\endgroup$ – Martin Argerami Feb 13 '12 at 5:16
  • $\begingroup$ @MartinArgerami: $L^2(\mathbb{R})$. $\endgroup$ – Antonio Vargas Feb 13 '12 at 5:33
  • $\begingroup$ In that case I would expect your case to be harder than those in the link you provided: because they require the kernel to be in $L^2$ of the product space, while yours isn't. $\endgroup$ – Martin Argerami Feb 13 '12 at 15:54
  • $\begingroup$ @MartinArgerami: Sorry, I should have been more clear, I meant $L^2([0,1])$. The kernel of the operator in question, $e^{xt}$, is certainly in $L^2([0,1]^2)$, and the operator acts on $L^2([0,1])$. The link definitely applies here. $\endgroup$ – Antonio Vargas Feb 13 '12 at 17:02
  • $\begingroup$ Just to clarify, is the eigenvalue problem defined as $$\int_0^1 K(x,t) g(t) dt= \lambda g(x)$$ where in your case $K(x,t)=e^{x t}$? $\endgroup$ – Yuriy S Oct 19 '18 at 13:10

We need to solve the eigenproblem:

$$\int_0^1 K(x,t) g(t) dt= \lambda g(x)\tag{1}$$

While I doubt that there's an exact solution in terms of known function, we can find an approximate one by exploiting the method described here for a certain class of kernels:

$$K(x,t)=\sum_{j=0}^n \phi_j(x) \psi_j(t)$$

The approximation is achieved by Taylor expansion of the exponential kernel:

$$e^{xt} \approx \sum_{j=0}^n \frac{x^j t^j}{j!}$$

From symmetry considerations, let us set:

$$ \phi_j(x)=\frac{x^j}{\sqrt{j!}} \\ \psi_j(t)=\frac{t^j}{\sqrt{j!}}$$

We need to compute:

$$a_{jk}=\int_0^1 \psi_j(t) \phi_k(t) dt=\frac{1}{\sqrt{j!k!}} \int_0^1 t^{j+k} dt=\frac{1}{(j+k+1)\sqrt{j!k!}}$$

Then the (approximate) eigenvalues are defined from the equation:

$$\det \{ a_{jk} - \lambda \delta_{jk} \}=0$$

In other words, we simply need to diagonalize the matrix $\{a_{jk}\}$.

The eigenfunctions will be:

$$g(x)=\sum_{j=0}^n C_j \phi_j(x)=\sum_{j=0}^n C_j \frac{x^j}{\sqrt{j!}}$$

Where $\{C_j\}$ are eigenvectors of the matrix.

As far as I see, the eigenfunctions for $n \to \infty$ should converge to the exact eigenfunctions, and in case they are analytical, the expansion above defines their Taylor series.

Here's a numerical check in Mathematica. For $n$ as small as $10$ we obtain a perfectly looking solution (the blue curve is the left hand side of the equation (1) and the orange dashed curve is the right hand side):

Nm = 10;
A = Table[N[1/(j + k + 1)/Sqrt[j! k!], 100], {j, 0, Nm}, {k, 0, Nm}];
{ln, Cn} = N[Eigensystem[A], 20];
g[n_, x_] := Sum[Cn[[n]][[j + 1]]/Sqrt[j!] x^j, {j, 0, Nm}];
Plot[{NIntegrate[Exp[x t] g[1, t], {t, 0, 1}], ln[[1]] g[1, x]}, {x, 
  0, 1}, PlotStyle -> {Thick, Dashed}]

enter image description here

And two more eigenfunctions:

enter image description here

enter image description here

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.