1: \begin{abstract}

2: We introduce a max-plus analogue of the Petrov-Galerkin finite

3: element method, to solve finite horizon deterministic

4: optimal control problems. The method relies on a 

5: max-plus variational formulation, and exploits the properties of projectors

6: on max-plus semimodules. 

7: We obtain a nonlinear discretized semigroup,

8: corresponding to a zero-sum two players game.

9: We give an error estimate of order $\sqrt{\Delta t}+\Delta x(\Delta t)^{-1}$,

10: for a subclass of problems in dimension 1. 

11: We compare our method with a max-plus based discretization method

12: previously introduced by Fleming and McEneaney.

13: \end{abstract}

14: