March 25, 2021
This is version 1.0.0 of Dymos. This release primarily removes some deprecated experimental features, along with implementing a few bug fixes.
- The RungeKutta Transcription is removed. #550
- Two-character location specifiers ('++', '--', '+-', '-+') are removed in favor of 'initial' and 'final' #556
- User must now specify
solve_segments='forward'orsolve_segments='backward'when using solve_segments capability (True is no longer valid). #557 add_input_parameterandadd_design_parameterdropped in favor ofadd_parameter. #558 #561- Removed the deprecated command line interface due to it being a somewhat hackish abuse of OpenMDAO's hooks. #563
- Removed the deprecated 'disc' node subset in favor of the more correct 'state_disc' #565
- Removed the deprecated 'custom_targets' option for parameters.' #565
- User will now be warned when multiple timeseries outputs were attempted with the same name. #567
- The
ode_classoption of phase is now marked non-recordable. #555 - Fixed doc building due to a change in OpenMDAO. #564
- Modified examples to use the preferred
fix_initial=Truefor time and states instead of "pinched bounds", e.g.initial_bounds=(0, 0). #560
February 18, 2021
This release of Dymos adds several examples demonstrating various capabilities of the code.
Per user request, the ODE of a system can now be provided via a callable function that
takes num_nodes and any other potential initialization keywod arguments.
This allows OpenMDAO's ExecComp to be used as an ODE if wrapped in a lambda, for instance.
Some bugs were fixed as part of introducing these examples. For instance, default values for states and times were being ignored - this is now fixed. In addition, Trajectory parameters are now saved to the simulation database file.
This is expected to be the final release of Dymos before v1.0.0, when several existing deprecated features will be removed from the code.
None
- Added the racecar example from @pwmdebuck, demonstrating cycling constraints and integration about arclength instead of time. #535
- Simplified the SSTO example to use a single ODE component with complex-step. #534
- Added a balanced field length example. #533
- Added the ability to use a callable that returns a System as an ODE. #528
- Removed duplication of inputs to timeseries when multiple outputs may use the same source data. #543
- Fixed a bug where timeseries outputs of non-dynamic ODE outputs would cause an exception. #521
- Reworked the cannonball example to make it more simple #545
- Placed some more tests under the use_tempdirs decorator to clean up output. #530
- Added a test that checks all docstrings vs the NumpyDoc standard. #526
- Clean up implementation of wildcard units when adding multiple timeseries at once. #523
January 21, 2021
This release of Dymos brings a few improvements and bug fixes.
We've implemented introspection for boundary and path constraints such that units and shapes of the constrained quantities can be determined during the setup process and no longer are required to be specified. The use of solve_segments to have a solver converge the dynamics and thus provide a shooting method at the optimizer has been improved. Option solve_segments can now take on values 'forward' or 'backward' to provide forward or backward shooting. Linkage constraints will now result in an error if the quantity on each side of the linkage is governed by the optimizer (is a design variable) but is fixed in value (not allowed to be varied).
- The command line interface is deprecated due to inconsistent behavior. #505
- Changed CoerceDesvar to handle vector values for ref/ref0/scaler/adder. Carried over changes to for defect ref/defect scaler. #464
- Changed solve_segments to allow it to be used when neither fix_initial nor fix_final are True. #490
- Added detection of unsatisfiable linkage constraints (where quantities on both sides cannot be changed by the optimizer). #502
- Added introspection for boundary and path constraints. #506
- Fixed a bug with using grid refinement if one or more of the states are vector-valued. #492
- Fixed issue where show_plots argument to run_problem is permanently changing matplotlib backend. #497
- Fixed a bug where trajectory.simulate() would crash if trajectory were not named the common 'traj' name. #500
- Fixed a bug that prevented restart from working with multiphase trajectories, and now include static parameters in the solution file. #510
- Fixed a bug associated with polynomial control rate sources, and added coverage of more cases. #513
- Fixed exception when simulating a trajectory with a parameter for which None was declared as the target for one of its phases. #515
None
December 14, 2020
This release of Dymos adds a few important features:
- Dymos can now automatically generate plots of all timeseries outputs
- State rates may now be tagged in the ODE
There are also several bug fixes.
- None
- Dymos can now automatically generate plots of all timeseries outputs. #469
- States can now be discovered automatically by tagging their rate variables in the ODE. #477 #481 #484
- Fixed issue where polynomial controls were not able to be linked across phases. #462
- Fixed a bug that was preventing parameters from being state rate sources. #466
- Users may now specify parameter shapes as integers or iterables other than tuples. #467
- Fix for apache license string in setup.py classifiers since it was not recognized by PyPI. #458
- Added a long description for PyPI. #460
- Some documentation cleanup for the JOSS review #474 #488
- Dropped the dependency on the parameterized package with the intent to utilize subTest in the future. #479
November 16, 2020
This release of Dymos fixes an issue that now allows portions of an array output to be connected to a parameter.
This version works with OpenMDAO 3.3.0 but version 3.4.1 offers some improved handling of parameters.
- Parameter shapes in 0.16.1 were stored as (1,) + the shape of the parameter. Now they are shaped as expected. #444
- State shapes and units, if not explicitly given, are now pulled from targets (if present and uniquely defined), or from the rate source variable. #449
-
Fixes a bug where user-defined shapes of states were colliding with those found during introspection, and other state introspection updates. #449
-
Fixes a bug that prevented the use of numpy arrays as a boundary constraints #450
- Added test to verify functionality of OpenMDAO 3.4.1 that allows the final value of a control (or a partial portion of any output) to be connected to a parameter. #445
October 23, 2020
This is a long-overdue release of Dymos that features several useful new features. By taking advantage of recent changes to the OpenMDAO setup process, we can use introspection to determine a lot of things automatically now. For instance, the user will no longer be required to specify the shape or default units of timeseries outputs or constraints (although they can override units if they choose).
Also on the topic of timeseries, dymos now supports glob patterns when adding ODE outputs to the timeseries. Adding all 100 outputs of a large ODE to the timeseries output can now be accomplished in a single line:
phase.add_timeseries_output('*')
Introspection is also used for states, time, and controls. States will get their default units and shape from the associated state rate variable now. Controls, Polynomial Controls, and Parameters will now get their default units and shapes from their targets.
With the advent of automatic IndepVarComps in OpenMDAO, there no longer needs to be a distinction between InputParameters and DesignParameters.
From here on out, users just add parameters to a Trajectory or Phase.
A user can choose to make them design variables (in which case they perform as DesignParameters), or not.
Connecting an external variable to the parameter makes it function as a InputParameter.
From here on, we plan on making releases of Dymos roughly once per month.
add_input_parameterandadd_design_parameterare deprecated and replaced withadd_parameter#365- The use of two-character location strings ('--' and '++') are replaced by 'initial' and 'final' when linking phases. #427
- The endpoint conditions component, used to pull out initial or final values of a variable in a phase, have been removed. Values should instead by pulled from the timeseries. #427
- The results of simulate are now automatically recorded to a database recorder 'dymos_simulation.db' and stored in a case named 'final'. Use
CaseReader('dymos_simulation.db').get_case('final')to access the case. instead of the previous behaviorCaseReader('dymos_simulation.db').get_case(-1)#437 - The RungeKutta transcription is deprecated. It is functionally equivalent to the GaussLobatto transcription with an order of 3, but since it's signficantly different under the hood it was being a drag on development.
One note on another backwards incompatibility. Since we're using some newer OpenMDAO capabilities for the way in which parameters are handled, it is currently not possible to connect an upstream output to a parameter in dymos when src_indices are used. We plan on fixing this issue with an upcoming OpenMDAO release. If you rely on this capability, we recommend waiting for that update before using this version of Dymos, or using an intermediate "pass-through" component to pull the correct index from the original output and then passing that as the value of a dymos parameter.
- Dymos components are no longer listed in check_partials output by default. Use
dymos.options['include_check_partials'] = Trueto override this for debugging. #438 - Any outputs can now be linked across phases. Added a more general and powerful
add_linkage_constraintmethod, but the simplerlink_phasesmethod remains in place for simple continuity. #422 - Time, controls, and polynomial control units are now determined from targets. #412
- States now pull their default units and shapes from the rate source. #398
- Phase method
add_timeseries_outputnow supports wildcards to allow multiple timeseries outputs to be added at once. #387, #387 - Timeseries outputs automatically detect shape and default units. #380
- Added ph-adaptive refinement method that is capable of shrinking the grid. #379
- Control targets will automatically be set to a top-level input of the ODE, if present. #373
- State targets will automatically be set to a top-level input of the ODE, if present. #356
- Water-powered rocket MDO example added. #343
- State rates are included in timeseries outputs by default. #329
- By default, run_problem will use a problem recorder to record only the 'final' solution after an optimization. #328
- Move most setup functionality to configure to allow more introspection changes. #327
- Include examples using IPOPT via pyoptsparse. #311
- Parameters may now be selectively omitted from the timeseries outputs, but are included by default. #305
- Added a test of the distributed ODE capability. #304
- Replace
load_casewithreinterpolate_solution. #296
- Fixed a bug that was causing matrix-shaped states to have connection errors. #441
- Fixed a bug where simulate failed when both standard and polynomial controls were present. #434
- Fixed an issue where parameters are promoted twice in some cases. This will cause errors in an upcoming versions of OpenMDAO. #429
- Fixed a bug that was causing matrix-shaped states to have connection errors. #441
- AnalysisError is now raised if scipy.integrate.solve_ivp fails during simulation. #400
- Fixed a bug involving GaussLobatto transcriptions
get_rate_sourcemethod. #334 - Fix for simulate encountering errors when time options
input_initialorinput_durationwere True. #317 - Fixed a bug in which shaped input parameters were breaking simulate. #301
- Sort linkage orders to make convergence more repeatable. #298
- Switch to use preferred
assert_near_equalmethod instead ofassert_rel_errorfrom OpenMDAO. #320 - Documentation is now handled via mkdocs instead of sphinx. #337, #332
2020-02-12
- [Bug Fix] Phase Linkage units now checks all optimal control variables from the first phase in the linkage for units. Previously units for things that were neither states, time, nor controls were not being assigned.
- [Enhancement] Removed Python 2 support.
- [Enhancement] Removed deprecated ODE Decorators.
- [Enhancement] Added ability to subclass Phase with an assigned ODE and default state variable options.
- [Docs] Added docs on subclassing phases.
- [Enhancement] Automated grid refinement is now available via the
dymos.run_problemfunction. - [Enhancement] Grid data is now available upon instantiation of the Transcription instead of being deferred to setup time.
- [Bug Fix] The user now gets a meaningful message if Phase.interpolate is called too soon.
- [Bug Fix] State rates are now correctly passed through the interleave component that provides timeseries outputs for Gauss-Lobatto transcription.
- [Enhancement] Added hypersensitive example problem.
- [Docs] Documentation added for grid refinement.
- [Enhancement] Deprecated the use of ODE decorators
- [Enhancement] Added shuttle reentry example problem
2019-07-18
- [Enhancement] Phase methods like
set_state_optionsandadd_controlno longer use **kwargs in order to make them more IDE friendly. - [Enhancement] Additional timeseries can for outputs can be added to a phase, with interpolation onto a new set of grid points. This enables the concept of tandem phases, where two different ODE's operating over the same time interval can be integrated on different grids. Fast state variables can be integrated on a dense grid while slower state variables are integrated on a more sparse grid, for performance. For an example see the tandem phase documentation in the feature docs.
- [Enhancement] ODE options can be specified at the phase level rather than in the ODE. This feature is experimental, but it allows one way of programmatically defining an ODE.
- [Enhancement] Changed the use of OpenMDAO to do
import openmdao.api as omfor consistency with the OpenMDAO documentation. - [Enhancement] Dymos is now imported in the examples as import
dymos as dm