#@ *: DakotaConfig=HAVE_NPSOL
#@ s6: TimeoutDelay=480

# DAKOTA Input File: dakota_pcbdo_short_column.in
# Optimization under uncertainty using polynomial chaos methods within a
# nested OUU environment.  The test problem is the "short_column" problem from
# Kuschel and Rackwitz, 1997.

environment,
	  method_pointer = 'OPTIM'

###########################
# begin opt specification #
###########################
method,
	id_method = 'OPTIM'
	model_pointer = 'OPTIM_M'
	npsol_sqp
	  convergence_tolerance = 1.e-6
	  output verbose

model,
	id_model = 'OPTIM_M'
	nested
	  variables_pointer  = 'OPTIM_V'
	  sub_method_pointer = 'UQ'
	  responses_pointer  = 'OPTIM_R'
# use projection of analytic PCE moments: constrain beta
	  primary_response_mapping   = 1. 0. 0. 0. 0.
	  secondary_response_mapping = 0. 0. 0. 0. 1.

variables,
	id_variables = 'OPTIM_V'
	continuous_design = 2
	  initial_point    10.    15.
	  lower_bounds      5.    15.
	  upper_bounds     15.    25.
	  descriptors      'b'    'h'

responses,
# minimize b*h
# s.t.     beta >= 2.5
# NOTE: This specifies the TOTAL RESPONSE for the optimization,
#       which is a combination of nested & interface responses.
	id_responses = 'OPTIM_R'
	objective_functions = 1
	nonlinear_inequality_constraints = 1
	nonlinear_inequality_lower_bounds = 2.5
	nonlinear_inequality_upper_bounds = 1.e+50
	analytic_gradients
	no_hessians

##########################
# begin UQ specification #
##########################
method,
	id_method = 'UQ'
	model_pointer = 'UQ_M'
	polynomial_chaos			#s0,#s1,#s4,#s5,#s8
#	stoch_collocation			#s2,#s3,#s6,#s7,#s9
	  askey non_nested
	  quadrature_order = 2			#s0,#s2,#s4,#s6
#	  sparse_grid_level = 1			#s1,#s3,#s5,#s7
#	  sparse_grid_level = 0			#s8,#s9
	  p_refinement
	    uniform				#s0,#s1,#s2,#s3
#	    dimension_adaptive sobol		#s4,#s5,#s6,#s7
#	    dimension_adaptive generalized	#s8,#s9
	    max_refinement_iterations  = 2	#s0,#s2
#	    max_refinement_iterations  = 5	#s1,#s3,#s4,#s5,#s6,#s7
	    convergence_tol = 1.e-4
	  num_response_levels = 0 1
	  response_levels = 0.0
	  compute reliabilities
	  cumulative distribution
#	  output debug

model,
	id_model = 'UQ_M'
	single
	  variables_pointer = 'UQ_V'
	  interface_pointer = 'UQ_I'
	  responses_pointer = 'UQ_R'

variables,
	id_variables = 'UQ_V'
	continuous_design = 2
	normal_uncertain = 2
	  means              =  500.0   2000.0
	  std_deviations     =  100.0    400.0
	  descriptors        =   'P'      'M'
	lognormal_uncertain = 1
       	  means             =  5.0
	  std_deviations    =  0.5
	  descriptors       =  'Y'
	uncertain_correlation_matrix =  1   0.5 0
					0.5 1   0
					0   0   1

interface,
	id_interface = 'UQ_I'
	direct
	  analysis_driver = 'short_column'
	  deactivate evaluation_cache restart_file

responses,
	id_responses = 'UQ_R'
	response_functions = 2
	analytic_gradients
	no_hessians