Consider the (compact) operator $T:C([0,1],\mathbb{R})\rightarrow C([0,1],\mathbb{R})$ s.t. \begin{equation} T(f)(x)=\int_0^1\min\{x,y\} f(y)dy \; . \end{equation} How could one find its spectrum? I have tried to impose $Tf-\lambda f \equiv 0$ but then I couldn't solve anything...

  • 3
    $\begingroup$ If you look for the keywords such as: operator min kernel covariance brownian, you will find a number of answers. E.g. here. $\endgroup$ – user66081 Aug 20 '16 at 5:56

A fundamental fact about this operator is that its inverse is the opposite of the second derivative. I show it in (Looking for examples of Discrete / Continuous complementary approaches).

What I give here is not an answer for the continuous operator "min" but for its discrete equivalent as a $n \times n$ matrix $M_n$ (or $M$) defined by

$$M_{i,j}=min(i,j) \ \ \ \ 1 \leq i,j \leq n.$$

Two reasons for studying this issue:

  • it helps to understand the behaviour of the continuous operator "min" .

  • it is essential in the implementation of this continuous operator on a digital computer.

In the same reference, I show also that the inverse of the discrete operator "min" as a $n \times n$ matrix still is the opposite $D_n$ of the discrete analog of the second derivative.

In my answer to a question (https://math.stackexchange.com/q/1880704), I show that the spectrum of $D_n$ is the set of

$$\lambda_p=4 \left(\sin\dfrac{p \pi}{n+1}\right)^2 \ \ (p=1 \cdots n)$$

Thus the spectrum of $M$ is the set of all $\dfrac{1}{\lambda_p}$.

Remark : $M_n$ and $D_n$, having positive eigenvalues, are symmetric positive definite matrices.

  • $\begingroup$ @William Tomblin You haven't made any comment on my answer... which was "alone in its category" :) $\endgroup$ – Jean Marie Jun 8 '17 at 0:46

It is fairly easy to solve the equation $Tf=\lambda f$ for $\lambda \neq 0$.. We have $\int_o ^{x} yf(y)dy +x\int_x^{1} f(y)dy=\lambda f(x)$. Observe that continuity of $f$ implies differentiability of the left side, hence of $f$ itself. In fact, by induction, we see that $f$ has derivatives of all orders. Differentiate once to get $xf(x)+\int_x^{1} f(y)dy-xf(x)=\lambda f'(x)$. Differentiate again to get $\lambda f''(x)=-f(x)$. Note that $\lambda$ is real because $f$ is real valued. The general solution of the equation $\lambda f''(x)+f(x)=0$ can be written as linear combination of sine and cosine functions if $\lambda >0$ and linear combination of $e^{cx}$ and $e^{cx}$, where $c=\frac 1 {\sqrt {\lambda}}$, if $\lambda <0$. Now remember that (from the steps leading to the differential equation) we have $f(0)=0$ and $f'(1)=0$. Now you can write exactly what the eigen values are and even write down the eigen functions.


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.