Refactored the Radau transcription to avoid the use of StateIndependentsComp #1117
Conversation
…ons as Radau.
Original Radau transcription can be accessed as RadauDeprecated, but will remain until the new one is thoroughly vetted.
RadauNew removes the need for the StateIndependentsComp which used a complicated hybrid IVC/implicit/explicit component implementation.
RadauNew uses the same structure as the Birkhoff transcription, where initial and final values of states in the phase are double-specified...once in the array of state values, and once as initial_states:{name} or final_states:{name}.
This provides a better source/target when linking phases together.
An important restriction of the new radau implementation is that it no longer supports compressed transcription. Continuity constraints are implemented between segments for state and control values.
…rently. Added scarring for t_final_targets that will not work until until that feature is added to dymos.
…DE using t_final_targets.
| @@ -66,7 +66,7 @@ def brachistochrone_min_time(transcription='gauss-lobatto', num_segments=8, tran | |||
| # Minimize time at the end of the phase | |||
| phase.add_objective('time', loc='final', scaler=10) | |||
|
|
|||
| p.model.linear_solver = om.DirectSolver() | |||
| p.model.linear_solver = om.DirectSolver(assemble_jac=False) | |||
There was a problem hiding this comment.
Does this test have problems with assembled-jac now?
| @@ -9,7 +10,7 @@ | |||
| OPT, OPTIMIZER = set_pyoptsparse_opt('SNOPT', fallback=True) | |||
|
|
|||
|
|
|||
| @use_tempdirs | |||
| # @use_tempdirs | |||
There was a problem hiding this comment.
Probably want to uncomment here and other places.
| @@ -1,5 +1,6 @@ | |||
| from .analytic.analytic import Analytic | |||
| from .explicit_shooting import ExplicitShooting | |||
| from .pseudospectral.gauss_lobatto import GaussLobatto | |||
| from .pseudospectral.radau_pseudospectral import Radau | |||
| from .pseudospectral.radau_pseudospectral import Radau as RadauDeprecated | |||
There was a problem hiding this comment.
I wonder if it would be better to rename the module to radau_deprecated, just to alleviate confusion when viewing the file in the future. Then radau_new can become radau_pseudospectral. Otherwise, kind of mixing paradigms here.
There was a problem hiding this comment.
While developing this I was planning on leaving it in, just in case some odd numerical issues cropped up. That doesn't seem to be the case in my testing, though. I'll likely just completely remove the old Radau implementation.
|
|
||
| def configure_timeseries_outputs(self, phase): | ||
| """ | ||
| Create connections from time series to all post-introspection sources. |
There was a problem hiding this comment.
nitpick, "to time series" and "from all post-introspection sources"
| """ | ||
| Declare group options. | ||
| """ | ||
| self.options.declare('grid_data', types=GridData, desc='Container object for grid info.') |
There was a problem hiding this comment.
Should this also be recordable=False?
There was a problem hiding this comment.
Actually not needed at all for that component. Removed.
|
|
||
| class RadauDefectComp(om.ExplicitComponent): | ||
| """ | ||
| Class definiton for the Collocationcomp. |
There was a problem hiding this comment.
Propagate rename through the docstring. -> RadauDefectComp
I think inclusion of initial and final defect are also new and should be mentioned?
…auNew now working with uncompressed transcription
…using compressed transcription.
…el testflo during debugging.
… provided by other state values. Documentation for tandem phase was not using the new set_xxx_val API.
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
…uNew. These are accessible through the Phase attributes ode_nonlinear_solver and ode_linear_solver, respectively. The solvers specified here are used by RadauNew in the ode_iter_group when solve_segments is used. Implicit duration now works with RadauNew, although we should change this to a more general implicit parameter capability in the future.
…t_defaults compatible shape issue addressed in OpenMDAO/OpenMDAO#3391
…in OpenMDAO reports.
…anscription might be applied to multiple phases while its solvers cannot.
…s enforcing linesearch is applied to phase.ode_nonlinear_solver. Radau aircraft case is working with IPOPT.
The original Radau transcription can be accessed as RadauDeprecated, but will remain until the new one is thoroughly vetted.
RadauNew removes the need for the StateIndependentsComp which used a complicated hybrid IVC/implicit/explicit component implementation. RadauNew uses the same structure as the Birkhoff transcription, where initial and final values of states in the phase are double-specified...once in the array of state values, and once as initial_states:{name} or final_states:{name}. This provides a better source/target when linking phases together.
An important restriction of the new radau implementation is that it no longer supports compressed transcription. Continuity constraints are implemented between segments for state and control values.
Summary
Summary of PR.
Related Issues
Backwards incompatibilities
None
New Dependencies
None