solution_level_cost

Cost estimates associated with a set of solution control values.

Specification

• Alias: None

• Arguments: REALLIST

Description

Simulation-based models may have an associated solution_level_control, which identifies a hierarchy of solution states, such as a set of mesh discretizations from coarse to fine, a set of solver tolerances from loose to tight, etc. In algorithms that manage such a hierarchy and perform optimal resource allocation among the solution states (e.g., multilevel Monte Carlo), it is important to estimate a set of costs associated with each state. These cost estimates can be relative, such as in the example below (lowest cost normalized to 1.)

Note: a scalar solution cost can be specified without an associated solution level control. This is useful when employing a hierarchy of model forms (each model has a scalar solution cost and no solution level control) instead of a hierarchy of discretization levels (one model has a vector-valued solution cost associated with multiple solution levels).

Examples

model,
 simulation
   solution_level_control = 'mesh_size'
   solution_level_cost = 1. 8. 64. 512. 4096.

variables,
 uniform_uncertain = 9
   lower_bounds      =  9*-1.
   upper_bounds      =  9* 1.
 discrete_state_set
   integer = 1
         set_values = 4 8 16 32 64
         descriptors = 'mesh_size'