Skip to content

v3.3.0

Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 11 Sep 14:14
· 122 commits to devel since this release
Immutable release. Only release title and notes can be modified.
v3.3.0
11aed33
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Changelog

Struphy 3.3.0 - 2026-09-11

Headlines

  • Enable automated scaling test workflow: running a profiling case across multiple MPI ranks, emitting per-run JSON metadata, packaging results into a publishable folder structure, and optionally uploading them to the struphy-hub/profiling-data repository. #266
  • New tutorials and verification tests for SPH: #261
    • tutorials/tutorial_viscous_euler_sph.ipynb
    • tutorials/tutorial_velocity_diffusion_sph.ipynb
    • tutorials/tutorial_hagen_poiseuille_sph.ipynb
    • tutorials/tutorial_dam_break_sph.ipynb
  • New guiding-center 5D phase-space conventions: switching from (v_\parallel, v_\perp) to (v_\parallel, \mu) as the standard coordinates in Particles5D. Use consistent mu_idx handling across particle kernels and PIC utilities. Introduced new CanonicalMaxwellian2D with caching capabilities. #275
  • New surface (boundary) integral operators in the FEEC layer: enable assembly/application of boundary mass operators needed for non-homogeneous derivative boundary conditions. #303
  • Enable parallel post processing on each rank. The default is still serial post-processing, which means that parallel_pproc=True needs to be passed for parallel work. #309
  • Added memory estimates in struphy and feectools: Adds per-rank memory footprint reporting for particles and key linear-operator types, including a dry_run construction mode for StencilMatrix to estimate allocation size without actually allocating. #363

API changes

  • Variables have been entirely removed from Propagator's Options classes. Such "Options"-variables were not updated by the Propagator, but nevertheless needed in the sub-step. From now on, such variables have to be passed to the Propagator constructor, which is done internally in the model file. The user cannot pass these variables anymore in the launch file through Options(). #262
  • Added a switch in EnvironmentOptions that lets you turn off writing of restart checkpoint data, this is convenient for benchmarking. #345

Physics models

  • New propagator PoissonAdiabaticGyrokinetic for the electric potential in gyrokinetic simulations with adiabatic electrons. #260
  • New examples for model DriftKineticElectrostaticAdiabatic: ITG in cylinderical geometry and cyclone base case (still issues due to noise). #260
  • New model for TAE energetic particle simulations: The new API now supports TAE energetic particle simulations with the model LinearMHDDriftkineticCC. #275
  • New model IncompressibleNavierStokesSPH: Incompressible Navier-Stokes equations solved using SPH with the Chorin projection scheme. The Poisson equation is solved on a grid with pressure in H1. This is a PIC scheme, not meshless anymore. #284
  • Three new verification tests for IncompressibleNavierStokesSPH (Chorin projection). #308
  • Added Poisson solver case to the profiling examples. #302

Bug fixes

  • Fix bug in mpi_sort_markers: add missing self.update_ghost_particles() to get the correct number of valid_mks. #261
  • Fix bug in matrix-matrix multiplication during weights assembling in WeightedMassOperators.create_weighted_mass(). #267
  • Fix incorrect accumulation kernels for basis_u == 1 and basis_u == v in drift-kinetic-MHD hybrid scheme. #352

User news and internals

  • For PIC, moved the factor 1/Np into the definition of the weights. #261
  • New files for sph kernels: eval_kernels_sph.py and pusher_kernels_sph.py. #261
  • Added a new Compiler class wrapping the existing struphy_compile CLI logic for API use. #281
  • Added a to_json() method to the Simulation class. #282
  • Added workflow to build a docs preview for existing PRs. #283
  • Added a new to_json() method to the Simulation class. #282
  • Added a new workflow to build a docs preview for existing PRs. #283
  • New accumulation kernel div_u_weak_1form for accumulating the divergence of the velocity field; new class ParticlesToGrid for passing the accumulation parameters to a propagator in the model init; and new logic in ImplicitDiffusion for passing accumulation and filter parameters, respectively. #284
  • Renamed propagator PushVinEfield -> PushVinForceField. #308
  • Use config.json instead of files in postprocessing. Instead of exporting all the classes as pickled binary objects, export the main information about the simulation as a json file called config.json. #314
  • Nightly automatic check of dependency bounds. #318
  • Enforce that feectools submodule should be pointing to the last commit. #320
  • Only allocate the full eval grid on rank 0. Grids are now rank local, _create_eval_grids returns global grids plus per-rank slices from derham.domain_array, so each rank holds only grids_log_loc. The physical grid is built on rank 0 only. #322
  • Added setup/modules.json for loading modules on known HPC systems. #326
  • Added setters to all properties of the Simulation class. #336
  • Added more profiling regions. #344
  • Fixed the xp.all(axis=1) bottleneck (10x speedup). #347
  • Column major fix of apply markers bc. Added a pre-allocated self._eta_bc_buf = xp.zeros((self.n_rows, 3)) buffer (1.3x speedup). #348
  • _sendrecv_get_destinations now checks neighbor ranks first, and only falls back to checking the rest if some markers remain unmatched (up to 1.5x speedup). #350
  • Updated profiling setup (scoper-profilerset to 0.5.0). Big improvements for GPU profiling and also line-by-line profiling for the regions we already defined using the tool without having to use line_profiler directly. #354 and #357
  • Enable setting log file with environment variable. #358
  • Improve docstring helpers. Added _html, _markdown, _latex methods to the model/domain/equilibria/perturbation baseclasses. The info(), pde(), ... methods remain the same, they just use the _html methods and display the code. #361
  • Split up docstrings for the Domain, Perturbation, and FluidEquilibria classes, mirroring the pattern already used by StruphyModel. #362

Struphy 3.2.0 - 2026-06-09

Headlines

  • New Quickstart, Userguide, Tutorials and Developer's guide in the documentation: #252
  • Each Propagator now sits in his own .py file: #236
  • New propagator CurlCurlSolve() for curl-curl problems: #245
  • New classmethods to inspect the model docstring (in a jupyter notebook for example): #229

API changes

  • Change in the signature of ParticleSpecies.set_markers(), which is used in parameter files featuring particles. The new classes SortingParameters and SavingParameters replace the methods set_sorting_boxes and set_save_data. Instances of the new classes are passed to set_markers(): #247

User news

  • Clean-up logging levels for simulation output: #247
  • New plotting functionality for kinetic backgrounds: #239
  • Addition of a matrix-free averaging operator for distributed FEEC data: #246
  • New default for boxes_per_dim is tuple = (1, 1, 1): #247

Bug fixes

  • Adaptation to general equilibria and new tests of the gyrokinetic Poisson solve: #238

Struphy 3.1.0 - 2026-04-24

Headlines

  • Refactoring of the Derham class: #212
    • Renamed Nel -> num_elements, p -> degree, nq_pr -> nquads_proj, polar_ck -> polar_splines
    • The type of spline basis functions is now set via the argument bcs, which is a tuple of length three. It holds the type for each direction: None for periodic, or a tuple of length two with either free or dirichlet to indicate the type of clamped splines. bcs replaces the two arguments spl_kind and dirichlet_bc
    • The lifting is defined for each FEECVariable separately through the new attribute lifting_function
  • Refactoring of WeightedMassOperators.create_weighted_mass method: #227
    • The signature remains largely the same, except that list input for weights has been replaced by tuple input. Moreover, within the tuple, one can now pass a SplineFunction object, either from H1 or L2 space, which will be multiplied to the constant weights during .assemble(). This allows for time dependent mass operators in nonlinear simulations (formerly treated with workarounds).
    • In the constructor of create_weighted_mass, we now evaluate all callables derived from weights: tuple[str] on the integration grid, multiply them together and then pass them as xp.arrays to WeightedMassOperator for building the object. This leads to much simpler code than in the previous version, where we built composed functions and passed them as weights.
    • In MassMatrixPreconditioner, to build the Kronecker matrices from 1D matrices without domain decomposition, the values of the weights are now retrieved via an MPI sub-communicator, because the weights in M0, M1 etc. are now given as local xp.arrays, not as callables anymore.
  • Use logging instead of print statements; use function set_logging_level from the API to set the logging level of all handlers. With pytest you can use pytest --logging-level DEBUG src/struphy to set the loggong level: #199 and #219
  • New user guide: #200
  • Remove two submodules struphy-parameter-files and struphy-tutorials in favor of the new folders examples/ or tutorials/: #206

API changes

  • base_units is removed from the StruphyModel constructor; all equation parameters that can be seen in the model docstring can be passed to the model constructor: #222

User news

  • Add to_dict and from_dict methods to the Simulation class: #186
  • Add __repr__ and __repr_no_defaults__ methods to most classes in the API: #193
  • Using pyvista; ddd show_3d and create_geometry_mesh methods to Domain class: #195
  • Add iterators to models and domains: #196
  • Added export and from_file methods to SimulationBase class: #197
  • Added name and description to the Simulation class: #198
  • New model ToyGyrokinetic to simulate the diocotron instability: #201
  • New classes for scalar quantities tracked during simulation (via new type Scalar): #220
  • The components of a model docstring are now available as class methods: #229

Bug fixes

  • Weights in initial Poisson solves of kinetic models without control variate fixed: #192

Struphy 3.0.4 - 2026-02-27

Bug fixes

  • New lower bound on pyccel is set to 2.2.0, due to this pyccel bug fix: pyccel/pyccel#2567

Struphy 3.0.3 - 2026-02-23

Headlines

  1. New class Simulation inherits the generic SimulationBase, both are in the new folder struphy/simulations/. The most important methods are:
    • Simulation.run()
    • Simulation.pproc()
    • Simulation.load_plotting_data()
    • Simulation.spawn_sister() (my new favorite!)

These are tested in the new tutorials: https://github.com/struphy-hub/struphy-tutorials/tree/use-species-properties

The file main.py has been deleted.

The Simulation takes a model as input. Other API classes are passed as well (see tutorials).
The model is viewed as everything related to the PDE, i.e. its variables, initial conditions etc. The simulation deals with the rest (geometry, derham, environment etc.)

Some important changes to the logic: The model does not have access to derham, mass_ops etc. anymore, these can be called from Propagator when needed. Solves that need to happen before the time stepping (like initial Poisson solves) are moved to model.allocate_helpers().

  1. The default launch file has been improved. See Tutorial 2 or test with struphy params MODEL. The files in the submodule struphy-parameter-files have been adapted.

  2. Several new classes have been introduced for post processing and plotting data, see post_processing_tools.py. The most important ones are PostProcessor and PlottingData. Dictionaries in the plotting data have been replaced by classes. Many classes now feature the __repr__ dunder for customized printing.

API changes

New classes exposed: Simulation, PostProcessor and PlottingData.

User news

  • Add set_zero_velocity argument into LoadingParameters, enforcing velocities of all particles along specified axis to always be zero: #176
  • New model ViscousEulerSPH replaces EulerSPH. The evaluation of the viscosity tensor has been implemented and tested for SPH methods. Unit tests for evaluation of the fluid velocity and its gradients (needed in the viscosity tensor) have been improved: #160

Struphy 3.0.2 - 2026-02-06

Headlines

  • Added a public API. This allows imports like from struphy import equils: #168
  • New default compile language is Fortran: #158
  • Moved each model to its own file. Calling sub-processes must be avoided in the future because of incompatibility with MPI: #152

User news

  • Added binning of higher order moments (current density, energy tensor) of f and delta f: #162

Developer news

  • Use pyccel 2.1: #153
  • Added three submodules: struphy-parameter-files, struphy-tutorials andfeectools. The Struphy repo should be cloned with git clone --recurse-submodules https://github.com/struphy-hub/struphy.git to init and update the submodules. Also, run git submodule update regularly to get updates from the submodules. See #154
  • Introduced class options.LiteralOptions for parsing literals. Moved Units to physics.py: #167

Bug fixes

  • Use struphy.io.options.Units in equils. This enables the use of GVEC, EQDSK and DESC in the new framework: #158

Struphy 3.0.1 - 2025-12-11

Headlines

Psydac is now installed via pip from our fork renamed to feectools, which is published on PyPI. This avoids installing from a .whl file.
Functionality remains unchanged, but future releases of Struphy are much easier and quicker. We have kept the option to install the upstream psydac, once it is also on PyPI.
See #147.

User news

None

Developer news

  • Removed legacy code (eigenvalue solver): #129
  • Add context manager to h5py.File() calls: #135
  • Fix undefined variables: #141

Bug fixes

  • Fix setter in DESCequilibirum, update quickstart guide: #132
  • Set defaults for given_in_basis: "0" for scalar and "v" for vector-valued: #136
  • Fix the restart function: #143

Struphy 3.0.0 - 2025-11-13

Headlines

Struphy 3 represents a major refactoring with breaking changes with respect to Struphy 2, in particular:

  • The .yml parameter files cannot be used anymore. Simulation parameters have to be transferred to the new .py launch files that are generated from struphy params MODEL. See the Struphy README for a quick introduction.
  • The console command struphy run ... has been deprecated. The new way to launch simulations is by executing the .py launch file, for instance with python params_MODEL.py.
  • Other deprecated console commands are struphy pproc and struphy units. Post-processing is now done through the API via main.pproc().
  • The Struphy repo has moved to Github. The old Gitlab repo will persist but not be maintained any longer. Issues, discussion and PRs will solely take place on the new Github repo.

User news

  • Please consult the Struphy README and links therein to get familiar with the new workflows.
  • New tutorials can be found on mybinder.

Developer news

Struphy has been refactored with the following principles in mind:

  • get rid of console commands and increase the use of the Struphy API wherever possible
  • become even more object-oriented
  • use Classes instead of dicts wherever possible
  • use Literals to show options for string arguments

In Struphy 3, models feature the following important objects:

  • ParticleSpecies, FieldSpecies, FluidSpecies

Each species is a collection of Variables:

  • PICVariable, FEECVariable, SPHVariable

These variables are updated by Propagators. All options for a simluation can be set in the new .py launch file.

Bug fixes

  • Incorporate psydac updates: #109
  • Auto install Psydac on first Struphy import: #118
  • Remove MPI Barrier responsible for deadlock: #121

Struphy 2.6.0 - 2025-11-12

Headlines

  • This is a test run for the relaease of Struphy 3.0 from the new Github repo

Struphy 2.5.0 and prior releases

Assets 3
Loading