1: \begin{abstract} 

2:   We propose a numerical method to solve the \MA equation which admits

3:   a classical convex solution. The \MA equation is reformulated into

4:   an equivalent first-order system. We adopt a novel reconstructed

5:   discontinuous approximation space which consists of piecewise

6:   irrotational polynomials. This space allows us to solve the first

7:   order system in two sequential steps. In the first step, we solve a

8:   nonlinear system to obtain the approximation to the gradient. A

9:   Newton iteration is adopted to handle the nonlinearity in the

10:   system. The approximation to the primitive variable is obtained from

11:   the approximate gradient by a trivial least squares finite element

12:   method in the second step. Numerical examples in both two dimensions

13:   and three dimensions are presented to show an optimal convergence

14:   rate in accuracy. It is interesting to observe that the

15:   approximate solution is piecewise convex in each element.

16:   Particularly, with the reconstructed approximation space, the

17:   proposed method demonstrates a remarkable robustness. The

18:   convergence of the Newton iteration does not rely on the initial

19:   values, which shows a very different behaviour from references. The

20:   dependence of the convergence on the penalty parameter in the

21:   discretization is also negligible, in comparison to the classical

22:   discontinuous approximation space.

23: 

24:   \noindent \textbf{keywords}: \MA equation, Least squares method,

25:   Reconstructed discontinuous approximation. 

26: \end{abstract}

27: