block:
ep:
   accuracy:
            stochastic:
                       order

   debug: 0
   memory: 0
   init: 0
   check_stability: 0
   lp: 5
   fp: 2
   innovation_distribution: 'gaussian'
                            'calibrated'
   maxit: 500
   periods: 200
   set_dynare_seed_to_default: 1

   solve_algo:
   stack_solve_algo: 4
   step: 50
   stochastic:
              IntegrationAlgorithm: 'Tensor-Gaussian-Quadrature'
                                    'Stroud-Cubature-3'
                                    'Stroud-Cubature-5'
                                    'Unscented'
              method: ''
              algo: 0
              order: 1
              hybrid_order: 0
              homotopic_steps: 1
              nodes: 3
              quadrature:
                         ortpol: 'hermite'
                         nodes: 5
                         pruned:
                                ortpol: 'hermite'
                                nodes: 5
                                pruned: [1x1 struct]
   verbosity: 0
initial_period: NaN
lmmcp:
      lb:
      ub:
      status: 0 (?? status is not an option ??)
mcppath:
        lb:
        ub:
minimal_solving_period:
order:
ramsey_policy:
simul:
      maxit
solve_algo:
stack_solve_algo:
ut: (unscented free parameter)

pfm.stochastic_order = DynareOptions.ep.stochastic.order;
pfm.periods = DynareOptions.ep.periods;
pfm.verbose = DynareOptions.ep.verbosity;


* in extended_path_core, one passes options.ep and individual options
* there are no options to control extended_path_homotopy
* extended_path_initialization sets ep and options
* setup_integration_nodes: number of nodes is not handled in a symmetric way for all algorithms
* why extended_path_initialization et setup_stochastic_extended_path ?
* do we need solve_stochastic_perfect_foresight_model.m and solve_stochastic_perfect_foresight_model_1.m ?