Version of the Day
Since the last major release (v4.2 on 11/5/08), the following
enhancements are available in the "version of the day" (VOTD)
releases. Stable release capabilities will typically lag VOTD
capabilities based on nightly testing results.
Version 4.2 (11/5/08)
Uncertainty Quantification (UQ) Capabilities:
- Stochastic collocation using Lagrange interpolation polynomials
is now supported for uncertainty quantification. See
Stochastic
collocation method and Section 6.4 of the v4.2 Users Manual for
additional information.
- Stochastic collocation (SC) and polynomial chaos expansions (PCE) now
support Smolyak sparse grids with mixed integration rules for more effective
scaling on problems with larger numbers of random variables.
- PCE/SC capability to numerically generate quadrature points/weights
in order to extend numerical integration capabilities beyond order 10 for
Hermite, Legendre, and Laguerre and beyond order 2 for generalized Laguerre
and Jacobi.
- Addition of PCE features supporting incremental studies and simplified
specification: (1) the point collocation option may now be specified using an
oversampling ratio, which eliminates the need for the user to calculate the
number of expansion terms each time, and (2) the point collocation and
expansion samples options may now reuse previous samples in the surrogate
construction.
- LHS now supports Gumbel and Frechet extreme value distributions.
- Random variable transformations (and associated Jacobians/Hessians
of these transformations) now support transformations from bounded normal
and bounded lognormal to standard normal and from triangular and loguniform
to uniform.
- Generalization of incremental random sampling in the Monte Carlo case
to allow for increments which are not factors of 2: increments of just one
or a few samples are now supported.
- Increased precision in output of response moments to enable convergence
plotting.
Optimization and Model Calibration Capabilities:
- Polynomial chaos-based design optimization now supports design
sensitivities of moments using either combined variable expansions of
response values ("all_variables" mode) or random variable expansions
of response derivatives (standard mode). In addition, support for
variable insertion (design variables which define distribution
parameters of random variables) and mixed design variable
insertion/augmentation has been added for both standard and
all_variables modes. See Section 11.6.2 of the v4.2 Users Manual for
additional information.
- Stochastic collocation-based design optimization has been added
and supports standard or all_variables expansion modes and inserted,
augmented, or mixed design variables.
- The surrogate_based_opt and surrogate_update
strategies are now the surrogate_based_local and
surrogate_based_global methods in order to consolidate
code with the efficient_global method and to allow
componentization of this code for usage elsewhere. As described in Surrogate-Based
Methods, all three methods (SBL, SBG, and EG) have been generalized
beyond surrogate-based optimization to additionally support surrogate-based
multiobjective optimization (MOO) and surrogate-based nonlinear least
squares (NLS), with options for the following approximate sub-problem
formulations: (1) "original_primary" poses the sub-problem using multiple
surrogate objectives or surrogate least squares terms for solution by a
MOO or NLS method, and (2) "single_objective" recasts the sub-problem to
use a single objective for solution by a standard single-objective
optimizer (allowing, e.g., the use of global optimizers for
surrogate-based MOO/NLS). The sub-problem minimizer may now be
specified by specification pointer (long form) or by name (short form).
- The surrogate-based global method now supports replacement or
appending of points in the surrogate construction iterations. A
goodness-of-fit test (currently the R-squared diagnostic) is also
used as a new termination control for the method.
- The hybrid optimization strategy has been refactored into a new
hierarchy
supporting the existing sequential hybrid as well as planned
new capabilities for embedded and collaborative hybrids. As part of this
development, the concurrent iterator parallelism level has been generalized
and is now supported for all strategies that contain iterator concurrency.
- Generalized the Pareto set strategy to include both multiobjective
optimization and weighted nonlinear least squares. Also enabled weighted
least squares and updating handling of primary response function weights
(least squares and multiobjective) throughout the Iterator hierarchy.
- NCSU Direct now supports batch evaluations, enabling parallel
executions. For coliny_direct, updated the default box division behavior
to better synchronize with parallel configuration.
- Direct interface to APPSPACK (independent of COLINY) allowing
linear constraint support and improved nonlinear constraint support.
- Addition of the capability to generate confidence intervals on
parameters estimated by the nonlinear least squares optimizers.
- Activated model calibration under uncertainty (MCUU) capabilities
based on sampling, reliability, and stochastic expansion approaches to
UQ. Added support for analytic sensitivities of response mean for MCUU
using the Mean Value reliability method to calibrate mean response values.
Framework Enhancements:
- "New IDR" (Dakota/packages/nidr) is now the sole input parser.
"Old IDR" had been deprecated but was supported in developmental
releases for an amount of time before being removed for the v4.2 DAKOTA
release. The new parser provides better error checking, lower
software maintenance requirements, and better scoping (so that
artificial keyword uniqueness requirements like the prepending of
variable type identifiers can be eliminated, and redundant keyword
specifications can be grouped). Backwards compatibility in input
files has been preserved through a new keyword aliasing capability.
- Boost
multi-index containers are now used for evaluation cache lookups.
Hash function lookups are much faster than the previous capability
(largely eliminating the need to "deactivate evaluation_cache" for
problems with high function evaluation counts). To restore the previous
list-based approach, configure the source distribution "--without-boost".
- Surrogates: addition of moving least squares and radial basis
function surrogate models to the Surfpack library. The Gaussian
process approximation now supports calculation of analytic derivatives
of the response surface with respect to the input variables (mean
trend only) for use by gradient-based iterative methods. Implemented
use of minimum vs. recommended samples across the hierarchy, where the
former is used for bounding and the latter is used for default values
(previously the minimum was the default). Addition of a flexible
mechanism for computing a variety of surrogate diagnostics (see Global
approximations). Refinement of sample reuse facilities: samples_file
specification (imported once at startup) is now handled separately from
reuse_samples (which may be applied on each approximate subproblem cycle).
- More complex analysis_driver syntax is now supported directly
without need to resort to analysis_usage (now removed). Verbatim
command lines (without parameters and results file name augmentation)
are also now optionally supported. Refer to Simulation interfaces
and Section 13.3.4 in the v4.2 Users Manual for additional information.
- Additional verbosity is available in NestedModel executions, for
variables/response scaling transformations, and for multipoint data
return from Optimizers.
- Encapsulation of nonlinear uncertain variable transformations and
interface to LHS within the new PECOS library.
- Migration from Epetra to Teuchos for serial dense vector/matrix
utilities and convenient C++ access to BLAS/LAPACK. UQ methods and
Gaussian process surrogates now employ Teuchos.
- Ability to use external GSL and/or Teuchos, from source or installed.
Also, use one copy of each throughout the DAKOTA build tree, including
sub-packages.
- Direct inclusion of AMPL in Dakota/packages instead of through Acro.
- All parser Data* objects upgraded to referenced-counted idioms
instead of copies to improve efficiency for large-scale problems.
- Major rework of variable mappings in NestedModels to support
additional mappings. Specified mappings are now performed using an
"All" view for the sub-model variables, allowing access to any of the
inner variables (including state). See Nested
Model Controls for additional information.
- Updates to the scaling/weighting machinery: corrected response
derivative scaling when variable scaling enabled.
Interfaces:
- MATLAB: improved documentation, support for newer 7.3/R2006b as
well as legacy MATLAB versions, configuration for Mac and Cygwin.
Improved handling of Matlab engine command (via environment variable
DAKOTA_MATLAB_ENGINE_CMD), Matlab structure persistence, and memory
management of Matlab fields. With thanks to Lee Peterson (CU Boulder).
- Library mode: support for use of DAKOTA as a service within another
application has expanded significantly. Three modes are now supported
for data input to DAKOTA as documented here,
allowing for more flexible specification of potentially large data sets.
To enable these three modes, input checking and default vector generation
routines have been extracted and generalized. The dakota_library_mode
executable has been added as a test capability for serial and parallel
plug-in interfaces (including support for passed-in MPI communicators)
and these three data input modes.
Miscellaneous:
- Generalization of distribution mechanisms to include internal-full,
internal-supervised, internal-web, and external distributions to allow
full nightly feature testing while supporting multiple third party
library constraints within developmental release distributions.
- Extensive rework of tutorial examples to assimilate templates
for input specification, simulation interfacing, and parallel
computing from a variety of sources (see Dakota/examples).
- Portability: Improved support for Darwin PPC, Cygwin, Intel and PGI
compilers. Updated installation guidance for Cygwin and Mac OS X.
Support of newer MPI distributions, including OpenMPI.
- Bug fixes: 66 Bugzilla ids (restricted access)
have been closed as part of the v4.2 development cycle.
Version 4.1 (9/21/07)
Uncertainty Quantification (UQ) Capabilities:
- New method: Wiener-Askey generalized polynomial chaos expansions
approximate the functional relationship between response metrics and their
underlying random variables. Orthogonal polynomial bases include Hermite,
Legendre, Laguerre, Jacobi, and generalized Laguerre. Expansion coefficients
are estimated by sampling, tensor-product quadrature, sparse grid cubature
(using Clenshaw-Curtis), or linear regression. Additional features include
analytic response moments, analytic random variable sensitivities, and analytic
design variable sensitivities. Refer to Section 6.4 of the Users Manual for additional information.
- New method: Efficient Global Reliability Analysis (EGRA) employs
mean and variance estimates from Gaussian Process (GP) models to form an
expected feasibility function (EFF), which is then maximized using a global
optimizer to select points for GP refinement. This process accurately
estimates response function contours, for which probabilities are then computed
using multimodal adaptive importance sampling. Refer to Section 6.3.3 of the
Users Manual for additional information.
- New method: incremental Latin Hypercube and Monte Carlo sampling.
In the incremental mode, one augments an initial LHS/MC sample set of size N
with a second sample set of size N. Both samples together (of size 2N) will
preserve the stratification and correlation structure of the original
sample.
- New method: importance sampling (standard, adaptive, and
multimodal adaptive) added for local and global reliability methods.
- Second-order integrations corrected (negative reliability indices and
Epetra symmetric matrix products) for use within local reliability
methods.
- All UQ methods extended to support generalized reliability indices.
- Epistemic methods now support final statistics results for use in
optimization under uncertainty (OUU). Refer to the Models chapter in the
Reference Manual for additional information on epistemic interval results
accessible within OUU and to Section 10.6.2 of the Users Manual for initial guidance on formulating
epistemic OUU studies.
- Added compile time switches for Tausworthe and Mersenne twister random
number generators for use with LHS/MC. These are not yet the default, as
performance testing is ongoing.
- Variance-based decomposition and Dempster-Shafer theory of evidence now
support multiple response functions.
Optimization Capabilities:
- The multilevel hybrid optimization strategy now supports flow of multiple
solutions between iterators. Iterators may now accept one or multiple starting
point and may output one or multiple final results. The strategy logic
maintains a consistent data flow and supports concurrency in multiple iterator
instances as required (i.e., multiple data points require multiple instances of
an iterator that accepts only a single input).
- the surrogate-based optimization (sbo) strategy now supports multipoint
approximations (i.e., tana-3), approximate subproblem reformulations for
nonlinearly-constrained problems (subproblem objective may be the original
surrogate objective, a surrogate lagrangian, or a surrogate augmented
lagrangian; subproblem constraints may be the original surrogate constraints,
linearized surrogate constraints, or no constraints), and constraint relaxation
using a heuristic homotopy approach. in addition, previous augmented
lagrangian updating approaches have been replaced with the provably-convergent
approach of conn, gould, and toint. see section 9.6 of the users manual for formulations and aiaa mao conference paper for
computational results.
- new strategy: a global surrogate-based optimization strategy has
been added. rather than being localized using trust-regions, each approximate
optimization cycle identifies points for continuing updates to a single global
surrogate model. this is most appropriate for use with interpolatory
surrogates, such as kriging/gaussian process models.
- new ouu approach: polynomial chaos-based design optimization
(pc-bdo) is now supported in a bi-level formulation employing one of three
different design sensitivity approaches: analytic design gradients based on a
pce formed over all variables, numerical design gradients based on a pce formed
only over the random variables, and numerical design gradients based on a pce
formed over all variables. in the cases where the expansion is formed over all
variables, only a single pce construction is required for the complete pc-bdo
process, whereas the expansions only over the random variables must be
recomputed for each change in design variables. refer to section 10.6.2 of the
users manual for additional information on design
sensitivity analysis formulations.
- New method: the NCSU implementation of DIRECT by Joerg Gablonsky
has been added as a global optimizer, augmenting the existing COLINY
implementation of DIRECT. Additional NCSU optimizers, especially implicit
filtering, will be added in time.
- New method: closely related to the new EGRA uncertainty
quantification method described above, the Efficient Global Optimization (EGO)
method has been added. It employs mean and variance estimates from Gaussian
Process (GP) models to form an expected improvement function (EIF), which is
then maximized using a global optimizer to select points for GP refinement.
This process balances the desires to exploit areas with good solutions and
explore areas with limited data, resulting in excellent global search
characteristics. This method supports nonlinear constraints through an
augmented Lagrangian merit function formulation.
- New method plug-ins: Support for dynamic optimizer plug-ins has
been added. This initially targets SNOPT and NPSOL as shared object libraries,
but will in time allow more flexible binding to commercial solvers at run time
(currently managed at configure/make time).
- OPT++ now supports an internal mode override, allowing more effective
combination of optimizer evaluation requests, as well as iteration reset
functionality, allowing an instance to be instantiated once and reused multiple
times.
- Update to the DAKOTA/AMPL interface to handle algebraic nonlinear
constraints using the AMPL specification. This update allows the use of
standard AMPL model files without modification of the constraint specification.
Nonlinear Least Squares Capabilities:
- Calculation of residuals can now be simplified through user specification
of a least squares data file, which provides the set of data to which a
computational model will be calibrated.
DACE Capabilities:
- Initial interface to PSUADE from LLNL, providing access to the Morris
One-At-a-Time (MOAT) method.
Framework Enhancements:
- DAKOTA input files no longer require the use of "\" for line continuations.
All examples have been updated to omit the backslash.
- Gaussian process surrogate models now support variance estimators for use
in expected improvement/expected feasibility functions; constant, linear, or
quadratic trend functions; global optimization methods for maximum likelihood
estimation; and point filtering.
- New derived Model class: RecastModel. This is a thin wrapper
around a sub-model allowing for linear and nonlinear recasting of its variables
and responses. Used by variable and response scaling, random variable
transformations, multiobjective optimization, merit functions (see SBO
discussion above), and expected improvement/feasibility, as described in
Section 10.3 of the Users Manual.
- Extensions to variables views in surrogate model construction. Global
surrogate classes previously assumed a hardwired "All" view (due to the use of
this view by DACE methods) and local/multipoint/hierarchical surrogates
previously assumed a hardwired "Distinct" view. Now general view support is
available in all surrogate types, which allows much broader surrogate
deployment than was previously possible. As part of this extension, sampling
methods now support additional sampling modes (ACTIVE, ACTIVE_UNIFORM, ALL,
ALL_UNIFORM, UNCERTAIN and UNCERTAIN_UNIFORM) and many classes have been
extended to support data translations among differing variables set views.
- Methods may now utilize mixed surrogate/truth models, as identified by a
surrogate identifier list. The initial use case is sequential RBDO.
- More flexible interface to input/output scaling and added support for
logarithmic scaling. In addition, numerical derivative estimation is now
performed within the scaled space to avoid step size thresholds.
- Extended instantiate "on-the-fly" support for Iterators, Models,
Interfaces, Approximations, and Constraints in to support advanced model
recursions in local/global reliability and PCE methods.
- Support for "NaN" and +/-"Inf" added to the input parser (for use in
failure recovery values) and the response file reader. For both cases,
mitigation approaches are iterator-specific. Currently, sampling methods will
omit samples that are not finite from their summary statistics. See Section
20.4 of the Users Manual for additional
information.
- New interfaces: a new direct interface to Python has been
developed and the existing direct interface to Matlab has been refined.
- Extended surrogate build/update/append functionality within Models,
Interfaces, and Approximations.
- Added representation access function for letter/envelope idiom to reduce
proliferation of virtual functions at envelope base classes.
Miscellaneous:
- DAKOTA now supports parallel make for concurrent compilation.
- Ports to MINGW for Windows are now supported by the autotools harness.
- Revision control for DAKOTA is now managed by subversion, allowing
directory reorganization/renaming, upgrade of our release processes to employ
release branches, and upgrade of our TPL processes to employ svn
externals.
- Stable releases have been added as new developmental releases (supplements
existing VOTD release mechanisms), and revision numbers are now embedded within
all developmental releases for traceability.
- The nightly test harness now includes coverage of restart capabilities, as
required for incremental sampling methods.
- Bug fixes (Bugzilla record id): reliability methods (#1327, #2799), SBO
(#2124, #2829, #2836), RBDO (#2828, #2832), PLplot window advance (#2180),
sampling bounds (#2824), COLINY interface (#2250, #2434), JEGA parallelism
(#1264), parallel usage of plug-in interfaces (#2826, #2827), parallel usage of
instantiate on-the-fly Models (#2834), label propagation through Model
recursions (#2831), IEEE special values in iostreams (#2855), derivative
estimation in recast models (#2835), library build dependencies (#2158, #2822,
#2843), and VOTD distribution process (#2841).
Version 4.0 (5/12/06)
Autotools
- Upgrade from Cygnus configure to the latest GNU autotools (e.g., autoconf,
automake). This renders the configure/make system much more scalable to
emerging platform build targets. It moves from a host/target-based approach to
a feature-based approach, where configure probes are used to detect available
compiler features. The 4.0 release supports our primary platform targets (Red
Hat, Solaris, AIX, IRIX, and OSF). Continuing refinements to the new
configuration system are planned, and future VOTD and interim releases will
support Mac OSX, Cygwin, and mingw, among others.
JAGUAR and AMPL
- A JAVA front end for DAKOTA (JAGUAR) is now available. At the time of
DAKOTA v4.0 release, JAGUAR is in 1.0 beta, but the basics of defining a DAKOTA
input file are available. Future enhancements for graphical simulation
interfacing and visualization of results are planned.
- As part of supporting the JAGUAR equation editor capability, DAKOTA now
supports an interface to the AMPL solver library. This allows for the use of
algebraic problem definitions to define any subset of an Interface's mapping of
Variables to Responses. The algebraic_mappings keyword specifies file
names that define AMPL expression graphs, and the AMPL solver library processes
the files to perform the algebraic mappings as well as evaluate any
derivatives.
Framework enhancements:
- Addition of "model" as a
sixth keyword specification block. This increases flexibility in model
recursion for local and hierarchical surrogates (the previous approach of
embedding the model recursion controls within the method specification was too
restricting since no method is used in the construction of these surrogates).
This specifically targets multifidelity and local Taylor-series based
surrogate-based optimization where the surrogate models involve additional
recursions (e.g., additional data fits or uncertainty analyses). This change is
backwards compatible for simple (i.e., single method) input files, but is
incompatible for more advanced multi-method input files such as SBO, OUU, RBDO,
etc.
- Migration of the previous approximation interface specification into the
new surrogate model specification. This allows for the consolidation of
previous layered model/approximation interface specification pairs (present in
many SBO input files) into new surrogate model specifications. As there is no
longer an application/approximation selection in the interface keyword, the
application keyword has also been removed.
- Active set generalization: application interfaces are now passed an
additional derivative values vector (DVV) to specify the required derivative
components. This is important for analytic gradients in optimization under
uncertainty (see RBDO item below), but unfortunately, also defines a change in
parameters file format which is not backwards compatible.
- Addition of analysis_components to system and fork application
interfaces. Again, this required extension of the parameters file format. Also
consolidated the previous modelcenter_file specification within this
new specification.
- Revised dependency management to allow broader usage of global utilities.
This particularly improves error handling in templated data types.
- Introduced the use of std::set<> and
std::map<>. With the removal of support for some old C++
compilers, these STL constructs have replaced many previous uses of
std::list<> for evaluation bookkeeping.
- Modified the creation mechanisms for
Iterator/Model/Variables/Interface/Responses objects to cache new object
creations and reuse them where appropriate. This reduces memory consumption in
multi-Iterator/multi-Model strategies, and also supports evaluation counter
persistence in reused Models and Interfaces. It does, however, require more
sophisticated management of parallel configurations within the Model classes
(since a single model may now be used in multiple parallel contexts).
Iterators and Strategies now manage evaluation counters with explicit control
over absolute or relative counts.
- Expanded the problem description database class into a class hierarchy,
with separation of generic parser operations and IDR-specific parser
operations. An XML database (using Xerces plus the JAGUAR XML Schema) is
envisioned for eventual addition to the hierarchy.
- Added support for a much finer grain of evaluation reporting. This is
needed for algorithmic performance analysis and is activated only for
output = verbose or debug. Value/gradient/Hessian
evaluation counters are used for each response function. This supplements the
previous counters which only track total/new calls to Interface::map() and do
not track the particular data requests for each call.
- Library mode extensions: addition of new model/interface access functions,
repair of model recursion access functions for altering envelope data, repair
of parallel configuration initialization for interface plug-ins, addition of
parallel configuration option for splitting intra-communicators down to a
single processor (COMM_SPLIT_TO_SINGLE), and development of a sample
interface plug-in class.
Iterative methods
- Update to JEGA v2.0, which adds convergence and niche pressure operators
for MOGA, improved and configurable user output, improvements to core data
structures, and numerous bug fixes.
- Update to the latest version of OPT++, which adds namespace support and
upgrades to NEWMAT11.
- Update to the latest version of DDACE, which adds support for
oa_lhs and one-way analysis of variance (ANOVA).
- SGOPT methods have been deprecated and replaced with new COLINY methods.
Most COLINY methods support general nonlinear constraints, and the COLINY GA
supports mixed integer design variables.
- Consolidation of specification (where possible) between JEGA and COLINY
genetic algorithms.
- All optimizers and least squares methods now support scaling in variables,
linear constraints, and nonlinear constraints. Scaling may be user-specified
characteristic value, by 2-sided bounds, or by 1-sided bounds.
- Second-order integrations, second-order limit state approximations,
two-point adaptive nonlinear limit state approximations (TANA-3), and
second-order mean estimations are now supported for both RIA and PMA
reliability methods. Second-order PMA uses a backtracking Newton solve to
update the PMA reliability target given a second-order probability target.
Hessians may be analytic, numerical, or based on quasi-Newton updates.
- The uncertain variable specification has been extended to include new
triangular, beta, gamma, Gumbel, Frechet, and interval distributions. The
reliability methods have been extended to support gamma, Gumbel, Frechet,
Weibull, uncorrelated loguniform, uncorrelated triangular, and uncorrelated
beta distributions, and the nondeterministic sampling methods have been
extended to include beta, gamma, and triangular distributions.
- An initial sampling-based implementation of Dempster-Shafer theory of
evidence is now available.
- An interface to NLPQLP has been added. Thanks to Bengt Abel at BMW for
providing the initial version.
- Granularity has been added to the virtual Iterator functions to support
additional abstract Iterator operations. The previous construct/run/destruct
progression has been extended to a construct/pre_run/run/post_run/destruct
progression.
Model components
- Previous versions of neural network, kriging, MARS, and regression
polynomials have been deprecated and replaced with the initial release of
Surfpack. Surfpack adds support for equality-constrained regression and will
provide radial basis function approximations in the future.
- Multipoint approximations are now supported in the Approximation hierarchy.
The Two-point Adaptive Nonlinearity Approximation (TANA-3) has been added and
may be invoked either as a general approximation method or as a special limit
state approximation selection within reliability analysis.
- Gaussian process global approximations are now supported in the
Approximation hierarchy. This class utilizes OPT++ to solve for the
approximation coefficients.
- Approximation classes extended to support an anchor point, a special data
point for which exact matching is enforced (e.g., using equality-constrained
least squares). Data requirement estimation is now more consistent, through
the estimation of the number of approximation coefficients, the number of
constraints, and the data content per sample. Improved efficiency of
SurrogateDataPoint lists by adopting a reference counting design and
eliminating unnecessary data copying.
- Migration of nonlinear constraint data from Iterator to Model in support of
strategies (e.g., new homotopy constraint-relaxation in surrogate-based
optimization) that modify nonlinear constraint bounds and targets. All
Iterators now pull constraint data at run time such that any Model updates are
properly captured.
- Variables objects have been generalized to utilize a pair of variables
views, one for the active view and another for the inactive view. This renders
the classes more general and more encapsulated from external logic.
- Response objects are now dynamically resized when DVV lengths are updated.
Response objects now support derivatives with respect to either active or
inactive continuous variables in a more direct and efficient manner.
Derivative estimation and management in Model has been generalized to support
dynamic derivative needs based on the current content of the DVV.
- Addition of a direct interface to the MATLAB engine. This currently
requires a dynamically linked DAKOTA executable.
Strategies
- Reliability-based design optimization (RBDO) now supports analytic
sensitivities of reliability, probability, and response levels with respect to
design variables that either augment the uncertain variables or define
distribution parameters for the uncertain variables. This enables
fully-analytic bi-level RBDO approaches.
- The new model recursion and sensitivity capabilities described above enable
a new capability for sequential RBDO using a trust-region surrogate-based
approach. In this approach, reliability estimates and sensitivities are used to
form surrogate contraints based on Taylor series. This approach is the most
computationally efficient RBDO approach available in DAKOTA to date.
First-order and quasi-Newton-based second-order Taylor series are currently
supported.
- The new model recursion facilities described above enable a new capability
for multifidelity surrogate-based optimization in which the low fidelity model
involves a data fit surrogate.
- The surrogate-based optimization strategy has been extended to support
iterate acceptance via a filter method, trust region ratio calculation via an
augmented Lagrangian merit function, constraint relaxation via a homotopy
method, and hard convergence assessment via nonnegative least squares Lagrange
multiplier estimation. Lawson & Hanson least squares routines have been
added for these Lagrange multiplier estimations. SBO with global data fits has
been restructured to allow hard convergence checks prior to surrogate
construction and passing of anchor points as embedded consistency constraints
(this allows the surrogate to be constructed once per cycle with embedded
consistency, rather than initially built, updated with the center, and then
corrected).
Packages and Utilities
- GSL updated to version 1.7.
- Epetra updated to version 3.4. This enables access to LAPACK serial dense
eigensolvers (used in second-order probability integrations).
- PLplot updated to version 5.5.2.
- DPrePro updated to support more flexibility in parameters file formats and
numerical field regular expression matching.
- Extensions to the Perl regression testing tool
(Dakota/test/sqa/dakota_test.perl) to support individual test extraction and
execution.
- A new Perl utility (Dakota/test/sqa/dakota_update_input.perl) is available
for upgrading old v3.x input files to v4.0.
- DAKOTA now supports a "dry run" mode, in which all parsing is performed and
all objects are instantiated, but the analysis is not performed. This is
selected using the new "-check" command line argument and is useful for
checking input files for errors prior to parallel queue submission.
Documentation
- The DAKOTA Users Manual has been migrated from FrameMaker to LaTeX and is
now managed in Dakota/docs. This simplifies the management of the Users Manual
and will promote more regular updating.
- SQE process documents updated to version 2.0.
- Addition of example files for utilizing multilevel parallelism with
concurrent mpiexec launches within a single PBS allocation.
Bug fixes
- Resolution of Bugzilla ids 1044, 1260, 1263, 1264, 1265, 1266, 1340, 1341,
1351, 1364, 1391, 1398, 1445, 1882, 1893, 1894, 1914, 1962, 1965, 1968, 2013,
2145, 2146, and 2161. A selected subset is listed below (emphasis on
resolution of reported 3.3 issues).
- Graphics windows and tabular data are now properly updated for Iterators
which employ nonblocking synchronization (requires data caching until
evaluation sequences are complete).
- Corrected bug when using secondary file tagging with fork interfaces.
- JEGA now properly configures processor allocations for message-passing
parallelism.
- Central composite and Box-Behnken designs of experiments, which are
inflexible in their supported sample size, now support improved logic for
managing user inputs, specification defaults, and minimum sample
requirements.
- population_size may now be specified independently from initialization_type
in COLINY and JEGA genetic algorithms.
- corrected SBO graphics and uninitialized additive correction arrays in the
case of a multiplicative to additive on-the-fly correction switch (performed
when numerical issues are encountered with multiplicative corrections).
- corrected "random" keyword duplication in initialization_type and
replacement_type for SGOPT/COLINY GAs.
- protected sample statistics calculations (standard deviations and
confidence intervals) for the case of only one sample.
Version 3.3 (12/23/04)
Framework enhancements:
- Revisions to the Iterator class hierarchy to bring together the
uncertainty quantification, parameter study, and design of experiments
classes so that common statistical analysis routines can be shared.
Iterator
is now divided into two branches,
"Minimizer"
(for optimization and nonlinear least squares) and
"Analyzer"
(for uncertainty quantification, parameter studies, and design of
experiments).
- Major redesign of multilevel parallelism classes to be more modular
and extensible (the number of parallelism levels is now open-ended)
and to allow multiple parallel configurations to coexist (previously,
only a sequence of partitions could be used, one at a time). This
development was driven by two needs: (1) performing communicator
partitioning at construction time instead of run time (for simplifying
parallel set-up for direct simulation interfaces), and (2) allowing more
complete parallelism management in strategies which have more than one
active model at a time (e.g., multifidelity SBO, see Strategies section
below). Features include:
- The new design uses an open-ended list of
ParallelLevel
objects which collectively contain all of the communicator
partitions. This list is managed with a list of
ParallelConfiguration
objects, each of which identifies the set of ParallelLevels which
define a particular multilevel parallel configuration.
- Additional encapsulation of message-passing routines through addition
of more MPI wrappers.
- Addition of a component parallel mode (INTERFACE, SUBMODEL, or none)
facility to the Model hierarchy to manage the Model component whose
parallel configuration is currently active.
- Addition of finite-difference numerical Hessians and BFGS/SR1
quasi-Newton Hessians to DAKOTA's derivative estimation routines.
Previously, DAKOTA relied on vendor optimization algorithms to perform
these calculations, but a native capability was needed for FD- and
quasi-second-order surrogate corrections. Features include:
- Numerical Hessians may use 1st-order gradient differencing or
2nd-order function-value differencing. This selection is
automatically determined based on the gradient specification: the
former if analytic gradients and the latter otherwise. FD Hessians
from 1st-order gradient differences have a symmetry enforcement
[H = 1/2 (H + H^T)]
- As for numerical gradients, the numerical Hessians support the
ability to be fully asynchronous (i.e., multiple gradient/Hessian
data sets may be evaluated concurrently, each of which involves
multiple finite difference evaluations).
- Quasi-Newton safeguarding: the BFGS update may be damped or undamped.
In the SR1 and undamped BFGS cases, the update is skipped if the
denominator is near zero.
- Quasi-Newton scaling: an initial scaling of y'y/y's I (Shanno & Phua)
is used for both BFGS and SR1 to accelerate the updating process.
- General support for variable insertion in nested models. This enables
(1) optimization under uncertainty, in which design variables are
distribution parameters, and (2) second-order probability, in which
epistemic uncertainties are modeled as intervals on distribution
parameters and realizations of these parameters are used within sets of
aleatory probabilistic analyses.
Nested models can update sub-model active/inactive variables using
any combination of variable insertions and
augmentations (see variable mappings in
Method Independent Controls).
- Extensions to DAKOTA's library mode. This mode has now stabilized
and is in use in several Sandia and external simulation frameworks.
It supports a "sandwich" implementation in which DAKOTA provides iteration
services to a simulation framework, and the simulation framework
provides a direct interface plug-in to DAKOTA.
The current state is reflected in Interfacing with DAKOTA as a Library.
Iterative methods
- Optimization:
- New Acro release with updated COLINY methods supporting general
nonlinear constraints: APPS, DIRECT, Solis-Wets, Pattern search,
and Cobyla. Updated DAKOTA input specification now captures the
relevant method options.
- The multiobjective genetic algorithm (MOGA) now returns the "best"
design to DAKOTA for use in multi-level strategies which require a
starting point for subsequent iteration. This best design is defined
to be the point in the Pareto set that has minimum distance from the
"utopia" point (the point of extreme values for each objective
function).
- Parameter estimation:
- Revision of NL2SOL specification mappings to be more consistent
with other methods.
- Uncertainty quantification:
- New capability for variance-based decomposition with the following
sampling methods: LHS, DDACE, and FSUDace. VBD provides global
sensitivity indices for each uncertain variable, where a higher
sensitivity index means that the uncertainty in that input has a
larger effect on the variance of the output.
- Addition of a 2nd-order probability capability as enabled by the
generic variable insertion capability described above. An epistemic
level can sample over uncertain variables, whose realizations are then
inserted into the distribution parameters for an uncertainty analysis
at an aleatory level.
- Addition of a subset of the GNU Scientific Library (GSL). In
particular, new routines for random number generation (the Tausworthe
RNG) and various statistical distributions have been added.
Confidence interval calculations for sample means and sample standard
deviations have been added to DAKOTA to take advantage of these
new capabilities.
- Improved logic in reliability methods for conditionally computing
response means/std deviations via MVFOSM analysis. The new
logic makes use of improved data management in nested models to
determine when these additional statistics are needed.
- Parameter studies/design of experiments:
- New FSUDace library for quasi-Monte Carlo (QMC) and Centroidal
Voronoi Tessellation (CVT) designs of experiments. The quasi-Monte
Carlo methods currently available are Hammersley and Halton sequences.
These new QMC and CVT methods support latinization of samples,
calculation of volumetric quality metrics, and variance-based
decomposition. This library was developed by Max Gunzburger and John
Burkardt at Florida State University.
- DDACE now supports volumetric quality metrics and variance-based
decomposition.
Strategies
- Surrogate-based optimization (SBO)
- Multifidelity SBO can now use more practical second-order corrections
based on the numerical and quasi-Newton Hessian capabilities discussed
above.
- Multifidelity SBO now supports parallelism in both the low and
high fidelity models, as enabled by the parallelism redesign
discussed above.
- Optimization under uncertainty (OUU)
- The generic variable insertion capability described above enables
reliability-based design optimization studies where the design
variables define distribution parameters (means, std deviations) for
uncertain variables.
- Nested models now compute a request vector for the sub-iterator
and support gradients/Hessians in their response mappings.
This enables sub-iterator sensitivities for accelerating nested RBDO.
Code for RIA/PMA reliability sensitivities has been fleshed out, but
is not active yet.
Miscellaneous
- Bug fixes:
- Extensions to IDR parser to provide more helpful error messages
in the case of unmatched keywords.
- Improved linear constraint management for mixed variable optimizers and
additional linear constraint error checking.
- Fixed discrete variable mappings in JEGA.
- Removal of static attributes from ProblemDescDB to correct
approach of multiple instantiations in library mode.
- Repaired long-standing memory leak in IDR. Static IDR table pointers
are now properly deallocated.
- Fixed bug with setup of random, chc, elitist selection in SGOPT GAs.
- SQA/Porting:
- Mac OSX port extended to include F90 LHS, MPICH, and BLAS linking
from g77.
- Initial support for g95 on Linux, OSX, Cygwin.
- Extensions to configure system to support --enable-debugging.
Version 3.2 (7/21/04)
Iterative methods
- Optimization:
- Initial release of JEGA: single-objective and multiobjective genetic
algorithms supporting mixed variables and general constraints. This
capability is in a beta release state (stable and undergoing minor
refinements).
- Initial release of COLINY: next generation of SGOPT including new
APPS and DIRECT algorithms. This capability is in an alpha release
state (still stabilizing).
- Updated versions of SGOPT, PICO, UTILIB, and OPT++. OPT++ version 2.2
adds limited-memory BFGS for quasi-Newton and disaggregated Hessian
nonlinear interior-point methods for constrained Gauss-Newton.
- Extensions to optimizers for improved tracking of final solutions in
multiobjective optimization problems. Single-objective optimizers,
multiobjective optimizers, and least squares solvers are now
standardized on reporting the full granularity of results (all
objectives or all least squares terms). In cases where a
single-objective optimizer is used to solve a multiobjective or
least squares problem (via an externally-applied multiobjective or
Gauss-Newton reduction), a database search is used to recover the
full granularity of data.
- Parameter estimation:
- New NL2SOL nonlinear least squares solver. This algorithm provides a
new capability for problems with large residuals at the solution (i.e.,
the Gauss-Newton approximation is inaccurate).
- Uncertainty quantification:
- New f90 LHS library now distributed with DAKOTA.
- New sampling method statistics: simple correlations (input-input,
input-output, and output-output) and partial correlations
(input-output) for raw and rank data.
- Extended reliability methods supporting the reliability index
approach (RIA) and the performance measure approach (PMA). These
approaches support the use of linearizations in x-space or u-space
centered at the mean values (x-/u-space AMV variants), iterated
linearizations in x-space or u-space centered at the most probable
point estimate (x-/u-space AMV+ variants), and no linearizations
(FORM variant). This increases the number of available reliability
methods from 4 to 11. The RIA approaches are used for computing
probability levels that correspond to user-selected response levels
(forward reliability analysis), and the PMA approaches are used for
computing the response levels that correspond to user-selected
probability levels (inverse reliability analysis). Additional
features include:
- support of NPSOL sequential quadratic programming or OPT++
nonlinear interior point methods for solving the RIA/PMA most
probable point searches. With the addition of OPT++ support,
DAKOTA's reliability methods are now available to users lacking
NPSOL licenses.
- analytic objective/constraint gradients are now supported within
all NPSOL/OPT++ RIA/PMA evaluation functions.
- warm starts are employed for the linearization point and
optimizer initial guess for the next AMV+ iteration increment,
for the next response/probability/reliability level, or for the
next design variable values. In some cases, projections are
used to avoid premature convergence (due to KKT condition
satisfaction).
- UQ standardization: consistent support for response (z),
probability (p), and reliability (beta) level specifications in all
sampling, reliability, and polynomial chaos methods. Mappings can
now be z->p, z->beta, p->z, or beta->z, and the z/p/beta levels may
now be mixed within the specifications for each response function.
Probabilities and reliabilities may now correspond to either cumulative
or complementary cumulative distribution functions (CDF/CCDF).
Specific extensions required for standardization include:
- the Mean Value reliability method now supports z/p/beta level
calculations (previously, only response mean and standard deviation
estimates were supported).
- MPP search-based reliability methods now support multiple response
functions and all z->p/beta and p/beta->z mappings (previously, only
z->p mappings for a single response function were supported). The
new z->beta mappings are especially important for RBDO since beta
is more well-behaved and well-scaled than p for use in optimization
constraints.
- sampling/PCE methods have added support for p->z, z->beta, and
beta->z for CDF/CCDF probabilities/reliabilities (previously, only
a single z->p CCDF mapping per response function was supported).
The reliability index beta is defined for sampling methods as the
number of standard deviations separating the mean from the target
response level. This gives an alternative metric for resolving
low probabilities with sampling methods for use in OUU.
- lognormal distribution parameters have now been standardized
within sampling and reliability methods.
- Parameter studies/design of experiments:
- Addition of a new base class, Dakota::PStudyDACE, which consolidates
previously replicated code between parameter study and design of
experiments iterators.
Strategies
- Surrogate-based optimization (SBO)
- New hybrid penalty/filter logic eliminates erroneous rejection of
SBO iterates. In particular, the penalty parameter is rescaled so as
to accept iterates with reductions in constraint violation (penalty
levels must increase monotonically). This is less restrictive than a
pure penalty approach but more restrictive than a pure filter approach.
A pure filter approach will be implemented soon.
- New 2nd-order additive and multiplicative local corrections and new
2nd-order combined/multipoint corrections have been added. Improves
convergence for SBO employing hierarchical surrogates.
- Enhancements to trust region ratio logic and soft/hard convergence
controls. The hard convergence check now calculates a projected
gradient norm in the presence of active bound constraints.
- SBO now overrides the default active set vector within the DACE model.
This is important when a specification of 1st/2nd derivative information
is needed to support higher order corrections, but the higher order data
is not needed for every DACE evaluation in a global surrogate build.
- The Taylor series approximation has been extended to automatically use
a 2nd-order series whenever analytic Hessians are specified in the
truth model response specification.
- Optimization under uncertainty (OUU)
- Efficiency of trust-region surrogate-based OUU algorithms has been
enhanced. In addition, lower-level approximations can now be
temporarily bypassed when performing "truth" verifications across
multiple surrogate levels.
- New bi-level reliability-based design optimization (RBDO) capabilities.
These RBDO methods employ warm started reliability analyses (initial
MPP search initial guess and linearization point).
- Model calibration under uncertainty: OUU formulations involving
least squares data sets are now supported in addition to
optimization data sets.
Facilities
- Utilities:
- Graphics extensions for better visualization of SBO and reliability
analysis results. Reliability analyses now provide CDF/CCDF plots,
which required 2D plotting extensions to admit general x-y plotting
(previously restricted to x axis counters), log scale for either/both
axes (only y axis supported previously), multiple data sets, and plot
markers with variable anchors.
- DPREPRO now supports assignments involving expressions and precedence
rules for duplicate assignments (allowing default template file
assignments to be overridden by parameters file values).
- Restart utility extended to support active/inactive variables,
response tags, sorting across multiple interfaces, and STL
iterator-based list traversals.
- DAKOTA library mode extended with improved support for publishing new
derived "letter" classes (instantiated outside of DAKOTA) to existing
"envelope" class hierarchies.
- General enhancements:
- New "Dakota" namespace shortens many DAKOTA class names and avoids
name clashes.
- utilib::PackBuffer replaced with Dakota::MPIPackBuffer. This
eliminates the dependence of DAKOTA MPI executables on UTILIB
which had been causing problems for AIX builds.
- Redesign and clean up of iterator classes employing function pointer
designs. Migrated from a design in which many static member attributes
were maintained solely as copies of nonstatic attributes to a design
in which these static member copies have been replaced by a single
static pointer to the active object instance of the iterator class.
This static pointer provides the needed data access from the static
member functions used in the function pointer design.
- IDR parser updated to use a factor of two buffer growth strategy.
This significantly reduces parsing time for large datasets.
- Addition of the ability to specify a vector of finite difference step
sizes, one per active continuous variable, for use in the DAKOTA
finite differencing routine (not supported by vendor routines).
- Bug fixes:
- NPSOL modified to prevent moving the initial point to the variable
bounds when the initial point is close to the bounds.
- Bug with number of correlation coefficients for one uncertain
variable corrected within new LHS sampling library.
- Bug fix in OPT++ within the line search backtracking routine
for the case of zero gradient.
- Optional response tag extraction is now more robust (would fail on
some platforms).
- Bug fix for parallel initialization of high fidelity interfaces in
model-hierarchy SBO.
- Bug fix for uninitialized pointers in the OPT++ interface classes.
- Bug fix in sampling output correlation matrix sizing for OUU.
- Clarification of best data point tracking in parameter studies and design
of experiments (only optimization and least squares data sets define best
data metrics, generic response function data sets do not).
- Make recursion bug fixed with addition of checks for missing vendor
optimizers/packages.
- SQA/Porting:
- Automated DAKOTA testing: nightly testing harness is much more thorough
and includes new unit test and parallel regression test facilities.
- Version of the day (VOTD) distributions are now active for publishing
the latest code versions to the user base.
- DAKOTA ports and configuration extensions: initial port to Mac OS X
completed; configuration support added for GM device of MPI, 64-bit
Sun builds, and Sun Cluster Tools MPI (/opt/SUNWhpc); Sun builds
updated to Workshop 7.
Version 3.1.2 (7/09/03)
Interim release enhancements
- Additional refinements to 2D graphics:
- rework of SciPlot numerics to correct problems that were occurring when the
y axis values become too large and/or when a range is not yet available:
- switched x and y axis numeric labels from a "%f" fixed format
to a "%g" output format which automatically switches to exponential
format when the numbers become large. This format also allows for
minimal label precision and field width.
- reworked numerics for calculating axis increments and estimating
space allocation for the new numeric label formats.
- reduced the maximum number of x axis increments to eliminate some label
overlapping and centered the x axis labels on the axis ticks.
- addition of plot window options for axis labels and legends.
- min/max symbols changed to unique shapes (discernable in black/white EPS).
- thumb wheels now support bidirectional windowing (left or right
and up or down).
- refinement of initial graphics window sizing for larger numbers of plots.
- elimination of some data skipping in the graphics through suppression
of data sets which contain only derivative data.
- efficiency enhancements through removal of redundant arrays and
streamlining of data updates.
- New features:
- initial release of
DPREPRO
(DAKOTA
pre-processor), a freely-distributable pre-processor
for use in interfaces between DAKOTA and various simulation codes.
DPREPRO is a Perl script utility that inserts values from the DAKOTA
parameters file into a template simulation input file to create new
simulation input file instances. Features include:
- supports both APREPRO and standard formats in the DAKOTA
parameters files and auto-detects the format in use.
- extracts variable assignments both from the parameters file and
from the template file and supports multiple order-independent
matches and replacements per line.
- stores tag/value pairs in a hash for lookup efficiency.
- supports algebraic manipulation using both the intrinsic Perl
functions as well as a variety of additional math subroutines.
- Miscellaneous:
- fixed a bug with management of the aprepro format for LayeredModels
which was occurring in hierarchical surrogate-based optimization.
- added a workaround for older pgf90 compilers such as those used on
ASCI Red to allow use of the new f90 LHS library within DAKOTA. To
mix C++ with pgf90, an explicit call to pghpf_init() is needed to
allow f90 calls to ALLOCATE().
Version 3.1.1 (5/21/03)
Interim release enhancements
- Overhaul of 2D graphics:
- subtle bug fixed with passing double* from an STL vector.
- horizontal/vertical wheel operations reworked.
- "Print" and "Markers" buttons activated in "Options"
window. Markers can now be deactivated, and individual EPS files
can now be generated from each plot window.
- box sizes, labels, button offsets, and colors adjusted to be more
legible, and initial window sizing adjusted to properly capture the
number of plots.
- reduced memory usage and improved efficiency.
- Completion of parallel scheduling for Pareto-set and Multi-start
strategies:
- fixed parallel bug in random start feature (LHS must be executed
only on world rank 0 and its results must be communicated to all
iterator masters).
- scheduling routines are more modular and dynamic scheduling is
now supported within dedicated-master partitions.
- Miscellaneous:
- added support for the GM device in MPI.
- streamlined parallel configuration reporting.
- fixed string output formatting problem under Linux.
Version 3.1 (4/04/03)
New iterative methods
- Optimization:
- NPSOL v5.0 update.
- OPT++ method consolidation (from 15 down to 6) and additional
constraint coverage (added BCFDNewton and GNewton NIPS).
- Initial rSQP++/MOOCHO interface (large-scale optimization).
- Initial COLINY interface for access to APPS and DIRECT optimizers.
- Uncertainty quantification:
- New f90 LHS library (not yet publicly available).
- restricted pairing: enforces the desired correlation in the sample
set, whether inducing a user-specified correlation or minimizing
correlation for the default uncorrelated case.
- bounded normal and bounded lognormal distributions
- bin-based and point-based histogram distributions
- additional output: sample histograms and achieved correlation in
f90 output, coefficient of variation and confidence interval
statistics in DAKOTA output.
- consolidation of old/new sampling for LHS/PCE
- MV/AMV/AMV+/FORM reorganization.
- Parameter estimation:
- New NLSSOL SQP-based generally-constrained nonlinear least squares.
- New least squares iterator branch to support NLSSOL and OPT++
GNewton.
- Sensitivity/statistical analysis:
- Distribution of DDACE under LGPL.
- Consolidation of function evaluation management routines among
design of experiments, UQ sampling, and parameter studies.
New strategies
- Surrogate-based optimization
- Enhancements to penalty function consistency.
- Enhancements to trust region management and convergence controls.
- New 1st-order additive correction (thanks to Natalia Alexandrov
and Michael Lewis for formulation). Improves convergence in SBO
test cases employing hierarchical surrogates.
- Optimization under uncertainty (OUU)
- New trust-region surrogate-based OUU
(paper).
- Enhanced data flow in Layered/Nested models.
- Concurrent (Pareto set, Multi-start)
- New random starts capability (in place of or in combination with
user-specified starts).
- New results/timing summaries.
- Parallel scheduling enhancements.
New DAKOTA facilities
- New library mode:
- Library creation (libdakota.a et al.) and consolidation in
Dakota/lib.
- New API for cases when command line arguments are not
directly accessible.
- Access to final strategy results.
- Library mode documentation.
- New surrogates:
- Linear and cubic polynomials to supplement existing quadratic
polynomial.
- Enhanced random seed control for multiple sampling sets in DDACE
and LHS
- 4 seed selections: combinations of fixed vs. sequenced and
user-specified vs. system-generated.
- Sequenced seeds result in variable sampling patterns which enhance
reliability and accelerate convergence in SBO, OUU, and SBOUU.
- New error checking in approximation classes, uncertain variables, PCE,
submodel compatibility, and method constraint support.
- Efficiency enhancements:
- Dropped pre-ANSI C++ mode which allowed for STL streamlining and
optimization.
- Additional output control: debug/verbose/normal/quiet/silent
settings, -i/-o/-e command line options.
- Evaluation cache and restart deactivation option.
- New MPI intra-communicators to streamline message broadcasts for
large parallel runs.
- Streamlined communicator repartitioning in strategies.
- New utilities:
- Restart repair by evaluation id.
- 3D graphics fonts (PLplot) distributed with executables.
- SQA:
- Automated DAKOTA testing: nightly build verifications, new regression
test facilities employing test matrices, unit testing under
development.
- Nightly library testing for SGOPT, PICO, and UTILIB (OPT++,
DDACE to follow).
- Requirements/issue tracking
(Bugzilla,
restricted access), design tracking, release management (automated
posting of nightly builds available shortly).
Contributions from the user community
- Distributed testing/issue identification:
- SGOPT: problem size limitations removed (identified by Michel
Beaulieu), real-valued GA bound constraint bug resolved (identified
by Hongman Kim), multi_step keyword mismatch resolved
(identified by Don Flaggs).
- OPT++: bound constraints fixed in PDS (identified by Don Flaggs),
final solution returned properly (identified by Michel Beaulieu).
- DOT/CONMIN: registration of automatic method switches for consistency
in output headers (identified by Don Flaggs).
- SBO fraction of cauchy decrease bug resolved (identified by Myles
Baker).
- LINUX graphics configuration generalized (identified by Constantine
Orogo).
- IBM POE: replaced short hand checks for MPI_REQUEST_NULL
(identified by Scott Brandon).
- DEC asynchronous evaluations: activated usleep() function
in asynch schedulers as OSF 5.1 is now supporting it (identified by
Scott Brandon).
- Enhancement suggestions:
- Surrogate builds from files (requested by Srinivas Kodiyalam)
- Output of surrogate functional forms (requested by Srinivas
Kodiyalam).
- Seed output from SGOPT to allow run recreation (requested by Don
Flaggs).
- Response function descriptors (requested by Hongman Kim).
- Contributed code:
- Cygwin evaluation and initial port (contributed by Hongman Kim).
- Mac OSX evaluation (contributed by Carl Hagelberg).
- Linear polynomial surrogate model (contributed by Myles Baker).
- Direct interface to Phoenix Integration's ModelCenter and merged
variable descriptors (contributed by Hongman Kim).
Version 3.0 (3/30/02)
All Version 3.0 beta additions plus:
General:
- Documentation (Users, Reference, and Developers Manuals) updated.
- New software quality assurance (SQA) standards instituted.
- Vector/list template classes based on STL are now well tested
and stable.
New strategies:
- The surrogate-based optimization (SBO) strategy reduces computational
expense and mitigates nonsmooth response variations through the use of local
(Taylor series), global (quadratic polynomials, neural networks, kriging,
splines), and hierarchical approximations. This strategy is now
provably-convergent through the addition of the "beta" correction approach,
which enforces first order consistency requirements necessary to ensure
convergence to the optimum of the actual, high fidelity model. This new beta
correction capability is critical for making SBO viable with hierarchical
surrogates (models of varying fidelity).
- The augmentation of samples at the root and branch nodes (for identifying
good NLP starting points) is now functional for the parallel branch and bound
strategy.
New iterative methods:
- OPT++ has completed the release approval process and the ANSI C++ version
of OPT++ is now available for download with DAKOTA. The performance of OPT++'s
nonlinear interior point methods is competitive with DOT and NPSOL, and OPT++
therefore provides a more capable freeware alternative for constrained
minimization than CONMIN.
- DAKOTA/UQ reorganized to make better use of inheritance, updated from Petra
1.0 to Epetra 2.0, and extended to include the mean value (MV) analytic
reliability method.
- DOT/CONMIN wrappers extended to manage linear inequality and equality
constraints. DOT, NPSOL, OPT++, and CONMIN are now completely plug and play in
terms of linear and nonlinear inequality and equality constraints.
New features:
- Tagged output streams for concurrent iterators are now managed with
Cout/Cerr macros and are independent of UTILIB's CommonIO class.
- Active set vector control may now be turned off in order to simplify user
simulation interfaces. This allows the user to neglect the active set vector
and trade some efficiency for simplicity.
- LayeredModel correction facilities streamlined to remove the need to back
out previous corrections before applying new ones. Local and hierarchical
layered models now operate correctly in parallel.
Version 3.0 beta (11/29/01)
General:
- ANSI C++ compliant: DAKOTA has been updated to the ANSI C++ standard.
Vector/list templates from the commercial Rogue Wave Tools.h++ library have
been replaced with the Standard Template Library (STL) provided as part of the
ANSI C++ standard.
- GNU GPL open source release: DAKOTA is now available for download
worldwide.
- New documentation: Users, Reference, and Developers Manuals now available
online from
http://www.cs.sandia.gov/dakota/documentation.html.
New strategies:
- Optimization under uncertainty (OUU): for robust and reliability-based
design in the presence of parameter uncertainty. OUU may be executed in simple
nested iteration mode, or it may employ surrogates at the optimization level,
the uncertainty quantification level, or both.
- Pareto-set optimization: generates a trade-off surface for multiobjective
optimization problems. The Pareto-set strategy supports concurrent-iterator
parallelism for large-scale multilevel parallel studies.
- Multi-start optimization: runs local optimizations from different starting
points for locating multiple minima. The multi-start strategy supports
concurrent-iterator parallelism for large-scale multilevel parallel
studies.
New iterative methods:
- The CONMIN library for gradient-based constrained minimization has been
added to provide an freely available alternative to the commercial DOT and
NPSOL libraries.
- Extensions to the DAKOTA/UQ uncertainty quantification library: stochastic
finite element techniques using polynomial chaos expansions have been added in
joint work with Johns Hopkins University.
New features:
- New tabular data output capability for postprocessing of DAKOTA results
with external post-processing tools such as Matlab and Tecplot.
- Graphics ported to SGI IRIX and PC LINUX platforms and extended for OUU,
multilevel hybrid, Pareto-set, and multi-start strategies.
