Problem definition#

A DMDO problem is described by a dictionary, usually read from a YAML file with load_setup() and built by Session (or build_mdo(), main()). This page lists every section and key. examples/basic_problem.yaml is a complete example (Getting started).

Important

Three settings have defaults that are easy to get wrong:

  • declare real variables with the type REAL. OMADS optimizes variables of any other type, including the short form R, on an integer lattice;

  • gamma defaults to 0: every link whose inconsistency is not exactly zero then counts as stalled and its penalty weight grows at every outer iteration. Values around 0.2–0.5 are usual;

  • scaling is used only when it is a float (0.1, not 10); otherwise the scale factors are computed from the variable bounds.

variables#

One entry per variable copy; each entry is a list:

variables:
  #    [name, subproblem, link, coupling, lb,  baseline, ub,  dim, type]
  v0:  ["u",  1,          2,    s,        0.,  1.,       10., 1,   REAL]
  v1:  ["v",  1,          None, u,        0.,  1.,       10., 1,   REAL]

Item

Meaning

name

Variable name. Copies of the same quantity in different subproblems share the name.

subproblem

Index of the subproblem that holds this copy.

link

Index of the subproblem holding the copy this one must agree with (a list for several), or None for a local variable.

coupling

s shared design variable, ff feed-forward coupling, fb feedback coupling, u local (uncoupled), CP constant parameter, dummy placeholder.

lb, ub

Bounds. They also set the default scale of the inconsistencies (Background).

baseline

Starting value.

dim

Dimension (vector-valued variables have one inconsistency per component).

type

REAL (recommended to give explicitly) or INTEGER. Default: R.

DA: disciplinary analyses#

Key

Meaning

index

Analysis index, referenced from MDA.

inputs

Variable keys the analysis reads, in the order the callable receives them.

outputs

Variable keys the analysis computes, in the order the callable returns them.

blackbox

Name of the analysis: a key of the session’s blackboxes mapping, a registered name, or an import path "package.module:attr".

type

callable.

links, coupling_type

Descriptive: the subproblem the outputs go to and the coupling type (ff, fb, s, un).

MDA: analyses per subproblem#

Key

Meaning

index

MDA index, referenced from the subproblem.

nAnalyses, analyses

Number and indices of the analyses evaluated, in order, for each design of the subproblem.

variables

Variable keys the analyses read from the subproblem.

responses

Variable keys the analyses compute.

coord: the coordinator#

Key

Default

Meaning

index

(required)

Coordinator index, referenced from the subproblems and MDO.

type

(required)

ADMM, the only coordinator type.

budget

(required)

Maximum number of outer iterations.

nsp

(required)

Number of subproblems.

index_of_master_SP

(required)

Subproblem that holds the system objective.

beta

1.8

Penalty growth factor \(\beta > 1\).

gamma

0.0

Stall threshold \(\gamma\): a link stalls when \(|q^{t}| > \gamma |q^{t-1}|\).

M_update_scheme

(required)

Which stalled links grow: normal, median, max or rank (Background).

scaling

bounds

A float scales every inconsistency by that constant; any other value uses \(0.1/(u-\ell)\) per copy.

mode

(required)

serial or parallel; the run’s mode argument takes precedence.

display, store_q_io

(required)

Console output of the coordinator; whether to store the inconsistency history.

adaptation_policy

none

Name of the coordination policy (Extending DMDO).

subproblem#

Key

Default

Meaning

index, nv

(required)

Subproblem index and number of design variables.

vars

(required)

Variable keys of the design vector x passed to the objective.

resps

(required)

Variable keys of the responses y passed to the objective.

is_main

(required)

True for the master subproblem.

MDA, coordinator

(required)

Indices of the subproblem’s MDA and coordinator.

opt, type

(required)

Name of the objective callable (resolved like blackbox) and callable.

solver

mads

Inner solver: mads, poll, search, scipy or a registered strategy.

budget

(required)

Evaluation budget of each inner solve.

scipy

SLSQP

For solver: scipy: {method, options, tol, is_con}. is_con: true passes the constraints to the optimizer; otherwise they are penalized.

configurations

none

Name of an entry of the CONFIG section with OMADS settings for this subproblem.

psize, pupdate

1.0, last

Initial OMADS frame size, and how it carries over between outer iterations (last, success, max, default).

fmin_nop

inf

Initial best objective.

freal

none

A reference value of the objective, used to report the relative objective error of the master subproblem.

display

False

Console output of the inner solver.

CONFIG: OMADS settings#

Named blocks of OMADS options, referenced by configurations. Each has an options part (seed, budget, tol, psize_init, opportunistic, rich_direction, check_cache, store_cache, …) and a search part (type, s_method, ns); see the OMADS documentation [Bayoumy, 2022] for their meaning.

CONFIG:
  CSP1:
    options: {seed: 12345, budget: 250, tol: 1.0e-9, psize_init: 2.0, opportunistic: False,
              check_cache: True, store_cache: True, rich_direction: True}
    search: {type: sampling, s_method: ACTIVE, ns: 10}

MDO: the run#

Key

Default

Meaning

coordinator, subproblems

(required)

Coordinator index and the list of subproblem indices, in sweep order.

responses

(required)

Variable keys of all analysis responses.

inc_stop

1e-9

Stop when the maximum inconsistency is below this value and no longer changes (must be a float).

noprogress_stop

100

Stop when the maximum inconsistency has not improved for this many outer iterations.

display

True

One progress line per outer iteration.

stop

"Iteration budget exhausted"

Initial stopping message, reported if no criterion triggers.

fmin, hmin, tab_inc

inf, inf, []

Initial values of the run’s records.

architecture

IDF

Informational; DMDO solves the distributed NHATC formulation of Background.

Sets and USER#

Sets defines named value sets for categorical and discrete variables. USER holds free-form user data, attached to the built problem.