1: \begin{abstract}

2: We consider the problem of reconstructing 

3: general solutions

4: to the Helmholtz equation $\Delta u+\lambda^2 u=0$,

5: for some fixed $\lambda>0$, on some domain $\Omega\subset \R^2$

6: from the data of these functions at scattered points

7: $x_1,\dots,x_n\subset \Omega$.

8: This problem typically arises when sampling acoustic 

9: fields with $n$ microphones for the purpose of

10: reconstructing this field over a region of interest $\Omega$ 

11: that is contained in a larger domain $D$ over which the acoustic field is generated.

12: In many applied settings, the boundary conditions

13: satisfied by the acoustic field on $\partial D$ are unknown

14: as well as the exact shape of $D$.

15: Our reconstruction method is based on

16: the approximation of a general solution $u$

17: by linear combinations of Fourier-Bessel functions or plane waves

18: $e_{\bk}(x):=e^{i \bk \cdot x}$ with $|\bk|=\lambda$.

19: We study two different ways of discretizing

20: the infinite dimensional space $V_\lambda$ 

21: of solutions to the Helmholtz equation, leading to two different

22: types of finite dimensional approximation subspaces, and we analyze the convergence of 

23: the least squares estimates to $u$ 

24: in these subspaces based

25: on the samples $(u(x_i))_{i=1,\dots,n}$. Our analysis describes the 

26: amount of regularization that is needed to 

27: guarantee the convergence of the least squares estimate

28: towards $u$, in terms of a condition that depends on the

29: dimension of the approximation subspace and

30: the sample size $n$. This condition also involves

31: the distribution of the samples and reveals the

32: advantage of using non-uniform distributions

33: that have more points near or on the boundary of $\Omega$.

34: Numerical illustrations show that our approach compares

35: favorably with reconstruction methods

36: using other basis functions, and other types

37: of regularization.

38: %

39: \end{abstract}

40: