1
$\begingroup$

I am trying to feed information about the solution when solving an inverse problem given by a Fredholm integral of the form $$ g(t)=\int_{a}^{b}K(t,s)f(s)ds. $$ Say I know $g(t)$ and $K(t,s)$, and want to know $f(s)$. But I have some additional knowledge: $g(t)$ and $f(s)$ are both constrained to the interval $[0;1]$.

Since this leads to an ill-posed problem, I wonder if the extra knowledge about the solution interval may help in finding a solution without struggling too much with inverse problem methods. Any help appreciated!

$\endgroup$
1

1 Answer 1

1
$\begingroup$

There is not enough information for a thorough answer. An a priori bound on the solution may indeed help theoretically and practically. As usual with measured data you may not want to solve the equation $Af=g$ (where $A$ denotes the integral operator, i.e $Af(t) = \int k(t,s) f(s) ds$) but a "least squares" type problem, i. e. $$\min_f \|Af-g\|. $$ You shall expect that minimizers do not exist (lack of coercivity). An a priori bound $0\leq f\leq 1$ can help as, depending on the function spaces, the problem $$\min_{0\leq f\leq 1} \|Af-g\|$$ may have minimizers.

From a practical point of view: Solving the constrained problem may indeed be easier but this depends on the specific situation.

As a first try you could give the projected gradient method a shot. Assuming that you work with the $L^2$ norm, this amounts to the iteration $$ f^{n+1} = P_{[0,1]}(f^n - \tau A^*(Af^n-g)) $$ with the adjoint $A^*$, a stepsize $0<\tau<2/\|A\|^2$ and the projection $P_{[0,1]} $ onto the interval, i. e. you apply pointwise $$ x\mapsto \min(\max(x, 0),1). $$

$\endgroup$
4
  • $\begingroup$ Thank you Dirk! I am indeed trying to use "least squares" i.e. minimization. Now, do you know any source of info about methods that make practical use of constraints on $g$ and $f$? $\endgroup$
    – Kurt
    Mar 16, 2015 at 20:03
  • $\begingroup$ The most simple thing that comes to mind is the projected gradient method, see my update. $\endgroup$
    – Dirk
    Mar 16, 2015 at 20:13
  • $\begingroup$ Is the method still valid if the unknown is $f$? If yes, I will definitely take a look at that! Thank you Dirk! $\endgroup$
    – Kurt
    Mar 17, 2015 at 12:39
  • $\begingroup$ For some reason I confused $g$ and $f$. Of course $f$ should be the unknown… Corrected. $\endgroup$
    – Dirk
    Mar 17, 2015 at 19:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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