Skip to content

Latest commit

 

History

History
231 lines (188 loc) · 8.91 KB

CHANGELOG.md

File metadata and controls

231 lines (188 loc) · 8.91 KB
Raw

CHANGELOG

version 3.0

Added

  • Sparse-dense Sylvester solver, see mess_sylvester_sparse_dense.
  • Krylov subspace projection methods for Lyapunov equations and AREs using extended and rational Krylov subspaces, see mess_KSM, as well as demonstration examples in KSM_FDM, KSM_Rail.
  • A new logging system allowing to print messages to both console and files, and switch between various output file format. See mess_log_*.
  • mess_tf_plot as unified backend for mess_sigma_plot, mess_Frobenius_TF_errorplot
  • mess_version a function to get the current version of M-M.E.S.S.

Changed

  • mess_galerkin_projection_acceleration has been replaced by mess_solve_projected_eqn, which now covers both the acceleration case and the case of projected equations in the context of mess_KSM.
  • eqn.st and eqn.nd from DAE usfs are now unified and called eqn.manifold_dim everywhere.
  • the LDL^T ADI for Lyapunov now expects W T W^T rather than G S G^T as the constant term. Consequently, eqn.S and eqn.G are now called eqn.T and eqn.W.
  • missing demonstration models are now loaded from the web on user request. The first run with a model may, thus, need to be interactive to confirm the download request.
  • usfs have been optimized further to not use unnecessary hidden data.
  • mess_lrnm, and mess_lrradi now require eqn.R and eqn.Q to be set in LDL^T mode.
  • operatormanager now has a mandatory pass-through argument opts for correct logging.

Fixed

  • many functions did not warn users about non-convergence
  • minor bugs in DRE methods were addressed
  • get_ritz_vals usfs and mess_para behavior has been unified
  • removed dead code in mess_lrnm
  • lyap in Octave and MATLAB had different behavior which is now properly wrapped to give consistent results
  • BDF methods of higher order now use lower order methods with smaller step sizes for a successive wind-up procedure to guarantee the expected order of convergence.
  • exact line search in mess_lrnm now uses a more expensive, yet far more numerically stable way to compute the relevant norms.

Deprecated

  • all functions in the morfolder are not going to be maintained in future releases in favor of using M-M.E.S.S. as the sparse solver backend for MORLAB in their version 6.0 and newer.

version 2.2

Added

  • Rail data are now available in additional sizes, via automatic downloads at runtime.

Changed

  • mess_balanced_truncation now adapts the default Lyapunov residual tolerances to the right hand side (RHS) and densifies RHS factors prior to compression.
  • Documentation has been updated especially for usfs sets

Fixed

  • mess_balanced_truncation would fail on certain incomplete opts structures, which are now reporting proper errors so users can adapt.
  • various smaller issues (outdated API, typographic errors) in the documentation have been fixed.
  • automatic downloading of additional examples now also works in Octave
  • mess_get_BIPS was failing on some models that used a differing naming scheme
  • some comments in the default usfs have been updated to better match the current implementation
  • figures now use consistent line-width for all graphs
  • typographic errors and outdated defaults in the help texts have been corrected

version 2.1

Added

  • Contribution guide & Code of Conduct
  • README file for the usfs available via help mess_usfs.
  • New solver for "Lyapunov plus positive" equations and related balancing based MOR of bilinear systems.
  • basic support for sparss and mechss system classes.

Changed

  • We changed the license to BSD 2-Clause!
  • README.md received some improvements pointing to references and documentation more explicitly.
  • Sigma magnitude plots now allow presampling of the full order model when, e.g. a sequence of ROMs is being processed.
  • mess_balanced_truncation now always performs a column compression of the right hand side factors in the two Gramian equations before starting the ADI.
  • mess_lradi does not require precomputed shifts anymore. It computes them automatically independent of the type now.
  • The RADI now allows for LDL^T factored right-hand sides in the Riccati equation and solution factors are expanded more efficiently.
  • LQR_FDM_unstable is now LQG_FDM_unstable_nwt and got fixed for appropriate computation of dual regulator and filter Riccati equations in ZZ^T and LDL^T formats. In this context, the new demo LQG_FDM_unstable_radi has been added to show the same computations using the RADI method.
  • Backend of mess_care changed from Newton method to RADI.
  • time recording in the demos is safer now.
  • dae_1_so usfs got rewritten completely and are documented much better now.
  • documentation of the second order usfs was improved.
  • the shift change criterion's default bound in tangential IRKA got lowered to 1e-6 (was 1e-2).

Fixed

  • mess_balanced_truncation would fail on certain incomplete opts structures, which are now filled automatically.
  • The TV2 model in the second order index-3 demo file had a typo.
  • LICENSE.md was showing a different license (fixed GPLv2) than the other files (GPLV2+)
  • dead code was removed or received better documentation of its purpose.
  • DRE solver received various minor fixes.
  • the tangential IRKA received some minor fixes, such that IRKA options are used more reliably and shift change monitory is more correct.
  • bugs in mess_lrradi for initial solutions and residuals in case of filter Riccati equation ('N') have been fixed.

version 2.0.1

Changed

  • many function headers and help texts got improved/completed

Fixed

  • DAE_1 usfs failed for certain systems with non-symmetric A.
  • LTV BDF could break in certain situations and was not following the general naming scheme for some variables.
  • mess_res2_norms would break when more than 4 output arguments were requested

version 2.0

Added

  • New RADI iteration for AREs
  • New splitting methods for autonomous DREs
  • New splitting and BDF methods for non-autonomous DREs
  • New operator manager only requires non-empty functions and replaces non-existent ones with a general mess_do_nothing function
  • renamed opts.bdf.stage to opts.bdf.step.
  • CI testing
    • demos serve as system tests
    • additional unit tests for the smaller building blocks and backend routines

Changed

  • improved Riccati iteration
  • updated minimum required/recommended Matlab and Octave versions (see DEPENDENCIES.md)
  • unified function interfaces for top level calls
  • unified handling of low-rank updated operators. Now always A+UV' is used. (Note the sign of the update and the transposition in V)
  • major updates in the MOR routines
  • some restructuring in the opts structure.
    • opts.adi.shifts has moved to opts.shifts such that also RADI can use it independent of ADI
    • opts.norm now determines the norm for all methods rather than having to consistently specify the same norm in each substructure
    • initial feedbacks for the Riccati solvers are now stored in the opts structure for the method rather than eqn
  • The projection shift routine uses the flag opts.shifts.implicitVtAV. Default is true. If set to false A*V is computed explicitly.
  • redesign of the demos
    • turned scripts into actual demo functions
    • new demos for indefinite AREs and H-infinity control

Fixed

  • several consistency updates and bug fixes
  • general code cleaning and pretty printing

version 1.0.1

Changed

  • updated documentation
  • Removed replacements directory since its content was not needed for Matlab after release 2010b and Octave after 4.0.

Fixed

  • Minor consistency and bug fixes and improved integrity of metafiles.
  • CI testing
    • demos serve as system tests
    • additional unit tests for the smaller building blocks and backend routines

version 1.0

Compared to the predecessor LyaPack a couple of things have changed.

  • The user supplied functions are now managed by an operator manager
  • The low-rank ADI now has:
    • optimized treatment of E matrices in generalized equations
    • more choices for shift selection, including completely automatic generation of shifts
    • improved stopping criteria based on low-rank factors of the current residual
    • automatic generation of real low-rank factors also for complex shifts
  • The Newton-Kleinman iteration features:
    • optimized treatment of E matrices in generalized equations
    • improved stopping criteria based on low-rank factors of the current residual
    • inexact Newton, line search and Galerkin projection acceleration
  • Examples have been extended
  • The Riccati iteration for H-infinity Riccati equations was added
  • DSPMR has not yet been ported to the new infrastructure
  • The SRM routine for balanced truncation is only available for none-DAE systems. Still, DAE versions are included in the corresponding DEMOS.
  • A tangential IRKA implementation for non-DAE systems was added