1: \begin{abstract}

2: We study the global convergence of the gradient descent method of the

3: minimization of strictly convex functionals on an open and bounded set of a

4: Hilbert space. Such results are unknown for this type of sets, unlike the

5: case of the entire Hilbert space. The proof of this convergence is based on

6: the classical contraction principle. Then, we use our result to establish a

7: general framework to numerically solve boundary value problems for

8: quasi-linear partial differential equations (PDEs) with noisy Cauchy data.

9: The procedure involves the use of Carleman weight functions to convexify a

10: cost functional arising from the given boundary value problem and thus to

11: ensure the convergence of the gradient descent method above. We prove the

12: global convergence of the method as the noise tends to 0. The convergence

13: rate is Lipschitz. Next, we apply this method to solve a highly nonlinear

14: and severely ill-posed coefficient inverse problem, which is the so-called

15: back scattering inverse problem. This problem has many real-world

16: applications. Numerical examples are presented.

17: \end{abstract}

18: