Numerical recovery of the 3D potential from fixed energy incomplete scattering data (Q918166)

Let \(\ell_ qu:=[\nabla^ 2+k^ 2-q(x)]u=0\) in \(R^ 3\), \(k=const>0\), \(q\in L^ 2(D)\) is real-valued, D is a finite domain, q vanishes outside D, \(u=\exp (ik\theta \cdot x)+A(\theta ',\theta)r^{-1}\exp (ikr)+o(r^{-1})\) as \(r:=| x| \to \infty,\quad xr^{-1}=\theta '.\) The inverse problem (IP) consists in finding q(x) from the knowledge of \(A(\theta ',\theta)\) known for all \(\theta ',\theta \in \tilde S^ 2\) and a fixed \(k>0\). Here \(\tilde S^ 2\) is an open set in \(S^ 2\), a solid angle. The first question is the uniqueness of the solution to IP. The second is the numerical construction of q. Proposition: The knowledge of \(A(\theta ',\theta)\forall \theta ',\theta \in \tilde S^ 2\) determines a real-valued compactly supported \(L^ 2\) potential uniquely. We discuss a numerical method for computing q(x) given \(A(\theta ',\theta)\forall \theta ',\theta \in \tilde S^ 2\). The proposed numerical method is the first mathematically justified method for solving the 3D inverse scattering problem with data given at a fixed energy. This inverse problem is highly ill-posed since small perturbations of \(A(\theta ',\theta)\) may result in a function which is not a scattering amplitude. Any numerical method for solving this inverse problem will be very sensitive towards noise in the data. This is an unavoidable feature of the problem. The solution can be stabilized if one assumes a priori more about q(x). The contents of this paper is to give a rigorous approach to solving the inverse problem with exact data, rather than to discuss the details of the numerical implementation of the method.