A stable numerical algorithm for identifying the solution of an inverse problem (Q2383747)

The article deals with the parabolic-type inverse problem \[ \begin{alignedat}{2} u_t&= (a(x) u_x)_x, &\quad\text{for } &0<x<1,\;t>0, \\ u( {x}, 0)&= f( {x}), &\quad &0<x<1, \\ u(0,t)&= p(t), &\quad &t>0,\\ u(1,t)&= \varphi(t), &\quad &t>0,\\ u(x_0,t)&= q(t), &\quad &0<x_0<1,\;t>0, \end{alignedat} \] posed for scalar functions \(u=u(x,t)\) and \(\varphi\), being given \(x_0\), \(q\), \(p\), \(f\) and \(a\), the latter assumed to belong to \(C^1[0,\infty)\) and to satisfy, for real constants \(A_0,a_0>0,\) \[ a_0\leq a(x)\leq A_0, \quad\text{for } 0<x<1. \] A finite difference algorithm is proposed, computer tests of its implementation being exhibited so as to back its claimed stability properties. This inverse problem may also be posed for a finite time interval, although the authors' uniqueness proof holds only for a half-line time span.