ESPResSo 4.2.0 released

This is a feature release, i.e., new functionality is added to ESPResSo. New thermostats, cell systems and boundary conditions have been introduced to simulate systems with Stokesian Dynamics, Brownian Dynamics, strongly inhomogeneous particle sizes or translation-invariant shear flow. The interface underwent (non-silent) changes, therefore scripts will have to be slightly adapted. Most notably, particle access by id and particle slices have a new syntax, and electrostatic/magnetostatic layer correction and reaction methods have a different setup. All errors are also now emitted as Python exceptions and are recoverable with minimal effort.

An additional focus of this release is the simplification of both the C++ core and the Python script interface to facilitate future extensions of ESPResSo. The testing of ESPResSo’s functionality has been extended considerably.

We recommend that this release be used for all production simulations. No further bug fix releases will be provided for the 4.1 line, and not all fixes are present in ESPResSo 4.1.4.

Please carefully read the detailed list of changes below before using this release. Issues can be reported at https://github.com/espressomd/espresso.

Get the source code in the Download area.

Added functionality

  • P3MGPU now supports energy and pressure calculation via the CPU kernels (#4506).
  • ELC now works with P3MGPU (#4506).
  • The LB grid now supports slicing operations (#4195) and LB slices are equality comparable (#4268).
  • Lees-Edwards boundary conditions can be used for particle-based simulations (#4457). Lattice-Boltzmann support will be added in the 4.3.0 release.
  • The non-bonded energy of a single particle can be calculated (#4401).
  • The list of close neighbors of a single particle can be extracted (#4401).
  • Brownian Dynamics simulations can be carried out with the newly added Brownian integrator and Brownian thermostat (#1842).
  • Stokesian Dynamics simulations can be carried out with the newly added Stokesian integrator and Stokesian thermostat (#3790, #3987).
  • Bonded interactions can now be automatically broken when the bond length exceeds a critical value (#4456). This feature can be combined with collision detection to model reversible bonds (#4464).
  • A new cell system HybridDecomposition was introduced to speed up simulations with inhomogeneous particle interaction ranges (#4373).
  • Shapes can be merged into meta-shapes (#3493, #3538).
  • The HollowConicalFrustum can now be sliced open, made thick and rotated to model quarter pipes in any orientation (#4179). The main application is in the construction of complex microchannel geometries via LBBoundaries.
  • A parametric weight function was added to the DPD interaction (#3570).
  • H5MD output files now support a unit system (#3751).
  • H5MD output files now support custom specifications to control which particle and box properties to write to disk (#4480).
  • The H5md class is now checkpointable and usable in an interactive Python session (#3751).
  • MDAnalysis integration now provides bond information (#3801).

Changed requirements

  • The minimal version of all dependencies was increased (#3375, #3687, #3878, #3984, #3994, #4115, #4312, #4337, #4489): Python 3.8, Cython 0.29.14, CMake 3.16, Boost 1.69, Sphinx 2.3.0, and Python packages versions are pinned on versions available in the Ubuntu 20.04 repository.
  • CMake no longer emits a warning about the deprecated distutils Python package, which is also no longer a requirement (#4433).
  • CUDA 11 support was added (#3870).
  • CUDA 8 and CUDA 9 support was removed (#3984).
  • AMD GPU support via ROCm (HCC and HIP-Clang compilers) was removed (#3966).
  • library libcuda is no longer a dependency in CUDA builds (#4095).
  • Installation instructions for ESPResSo on Microsoft Windows via WSL are now available (#4348).
  • LaTeX is no longer a requirement for building the Sphinx documentation and running the tutorials (#3256, #3395).

Feature configuration at compile time

  • GPU support is now opt-in (#3582). Pass the CMake flags -DWITH_CUDA=ON to compile CUDA code and optionally -DWITH_CUDA_COMPILER=<compiler> to select the CUDA compiler: NVCC (default), Clang.
  • Optional features HDF5, ScaFaCoS and Stokesian Dynamics are now opt-in (#3735, #4112). If they are requested with their -DWITH_<FEATURE>=ON flag and their dependencies are not found, CMake will raise an error. In the older 4.1 build system, CMake would silently ignore these features when their dependencies were not found, causing confusion as to what was exactly compiled.
  • Experimental support for fast-math mode was added (#4318). Some features might break depending on the compiler used to build ESPResSo. Please quantify the numerical stability of your simulations before enabling fast-math mode in production.
  • The LANGEVIN_PER_PARTICLE feature was renamed to THERMOSTAT_PER_PARTICLE (#4057).
  • The magnetostatic extension DLC now depends on feature DIPOLES instead of DP3M, since FFTW is not a dependency of DLC (#4238).
  • The electrostatic extension ICC now depends on feature ELECTROSTATICS instead of P3M, since FFTW is not a dependency of ICC (#4238).
  • The MMM1D_MACHINE_PREC feature was added to enable Chebychev series for MMM1D on CPU without the need to define the (now removed) BESSEL_MACHINE_PREC macro (#4311).
  • The EXPERIMENTAL_FEATURES feature was removed (#4482).

Improved documentation

  • Tutorials have been renamed and organized by difficulty level (#3993).
  • Tutorials Lennard-Jones, electrostatics, lattice-Boltzmann, raspberry electrophoresis and constant-pH have been improved (#3408, #3881, #3914, #3893, #4302, #4262).
  • Tutorial lattice-Boltzmann was split into three tutorials: polymer diffusion, Brownian motion and Poiseuille flow (#4052, #4329).
  • The active matter tutorial was rewritten into a Jupyter notebook (#3395, #4304).
  • An error analysis tutorial was added (#4174).
  • Tutorials now use the exercise2 plugin to hide solutions (#3872); since this plugin only exists for the classic Jupyter Notebook, a conversion script is provided for JupyterLab users (#4522).
  • The user guide now includes a button on Python code samples to hide terminal output and Python prompt symbols (>>> and ...), so as to facilitate copy-pasting examples directly in the terminal (#4386).
  • The user guide now uses a responsive theme for mobile/tablet users (#4504).
  • The user guide chapter on thermostats was moved to the chapter on integrators, since they are tightly coupled (#4080).
  • Mentions to non-existent functions were removed from the user guide (#4482).
  • Scientific publications referenced in comment lines in the core have been converted to BibTeX citations and integrated into Doxygen blocks to make them accessible in the Doxygen HTML documentation (#3304).
  • The Reaction Field electrostatic method is now documented (#4218).
  • The H5MD feature is now better documented (#4480).
  • A Gibbs ensemble sample was added to simulate the exchange of particles between two ESPResSo systems via the multiprocessing Python module (#4243).
  • A reaction ensemble sample was added to simulate a complex chemical reaction involving 5 chemical species (#3778).

Interface changes

  • The system.set_random_state_PRNG() method was removed (#3482).
  • The []-operator on system.part was removed (#4402). Use system.part.by_id(1) to fetch a specific particle, system.part.by_id([1, 3]) to fetch a group of particles, or system.part.all() to fetch all particles. This change was necessary to resolve the ambiguity of particle slices containing non-contiguous particle ids.
  • The domain decomposition cell system was renamed to regular decomposition (#4442). The system.cell_system.set_domain_decomposition() function was renamed to system.cell_system.set_regular_decomposition().
  • Bonds are now immutable (#4350). Bonds added to the list of bonds can no longer be overwritten by a bond of a different type, as it could lead to undefined behavior when the number of bonded partners was higher in the overwriting bond. Bonds can now be removed from the list of bonds, after they have been removed from particles.
  • Observable parameters are now immutable (#4206, #4211).
  • The Electrokinetics actor parameters are now immutable (#4327).
  • The LBFluid, LBFluidGPU, Electrokinetics and Species methods print_*() have been renamed to write_*() (#4049).
  • The ELC actor is no longer an electrostatics extension (#4125, #4506). The ELC actor now takes a P3M or a P3MGPU actor as argument and modifies it. Only the ELC actor needs to be added to the system list of actors. The ELC actor can now be removed from the list of actors.
  • The DLC actor is no longer a magnetostatic extension (#4506). The DLC actor now takes a magnetostatic actor as argument and modifies it. Only the DLC actor needs to be added to the system list of actors.
  • The NpT thermostat now uses the Philox random number generator and requires a random seed on first instantiation (#3444).
  • The analysis module energy() function now returns the lower triangle of the non-bonded interaction matrix, to be consistent with pressure() and stress_tensor() (#3712).
  • The analysis module energy(), pressure() and pressure_tensor() functions now return only two slots for electrostatics and magnetostatics: short-range contribution in the first slot and long-range contribution + layer correction in the second slot (#3770).
  • The analysis module pressure() and pressure_tensor() functions no longer provide a velocity-compensation flag to compute the pressure at half the time step in NpT simulations (#3756).
  • The espressomd.reaction_ensemble module was renamed to espressomd.reaction_methods (#4482).
  • The argument temperature in reaction methods was renamed to kT for clarity (#4305).
  • All reaction methods now take keyword arguments instead of positional arguments (#4451).
  • The constant pH method now implements a symmetric proposal probability instead of an asymmetric proposal probability (#4207).
  • The reaction method parameter exclusion_radius was renamed to exclusion_range (#4469).
  • Reaction method now take an optional parameter exclusion_radius_per_type for better control of the exclusion radius in simulations involving different particle sizes (#4469).
  • The WidomInsertion.measure_excess_chemical_potential() method was replaced by WidomInsertion.calculate_particle_insertion_potential_energy(), which returns the instantaneous value of the excess chemical potential instead of the accumulated mean and standard error (#4374). The mean value and standard error of the excess chemical potential must be now be calculated by WidomInsertion.calculate_excess_chemical_potential().
  • Reaction method constraints can now be safely changed from cylindrical to slab and can be removed (#4310). They will also raise an error when created with invalid parameters.
  • The mpiio global variable was removed (#4455). The MPI-IO feature is now used by creating a local instance with mpiio = espressomd.io.mpiio.Mpiio().
  • The MPI-IO feature now raises an exception from which the user can recover when the simulation script runs on 1 MPI rank, instead of an unrecoverable fatal error (#4455). This change is meant to help debugging read/write errors in simulation scripts. On 2 or more MPI ranks, exceptions still lead to a fatal error.
  • The H5md class takes new arguments during instantiation (#3785).
  • The system.cell_system.get_pairs_() method was renamed to system.cell_system.get_pairs() and now supports filtering particle pairs by type (#4035).
  • The polymer setup code was moved from the core to Python (#3477). The espressomd.polymer.positions() function was renamed to espressomd.polymer.linear_polymer_positions() and the espressomd.diamond.Diamond class was converted to function espressomd.polymer.setup_diamond_polymer(). For diamond polymers, counter-ions must now be added manually by the user.
  • The particle director can be set from the Python interface (#4053).
  • The particle method vs_auto_relate_to() can take a particle as argument instead of a particle id (#4058).
  • Particles can be serialized and deserialized in the Python interface with particle_dict = p.to_dict() and system.part.add(particle_dict) (#4060).
  • It is no longer necessary to manually reshape the output of Observable objects. The Observable classes now return multi-dimensional numpy arrays and the documentation clearly indicates the shape and size of the calculated data (#3560). The same applies to accumulators and time series (#3578).
  • Accumulator and Correlator classes now return the data in suitably shaped multi-dimensional numpy arrays; dependent properties such as lag times and sample sizes need to be obtained separately by calling methods lag_times() resp. sample_sizes() (#3848).
  • Profile observables provide methods bin_centers() and bin_edges() to facilitate plotting (#3608).
  • The observable ComForce was renamed to TotalForce, so as to better reflect what it actually calculates (#3471).
  • The RDF feature was removed from the analysis module and converted to an Observable class (#3706). Time averages can be obtained using the TimeSeries accumulator.
  • All occurrences of “Stress Tensor” in the analysis module, LB module and EK module were renamed to “Pressure Tensor” to better reflect what is actually calculated (#3723, #4228).
  • The MeanVarianceCalculator interface changed (#3996).
  • Observables now check their input parameters (#4211, #4255) and raise an exception when an invalid value is detected (e.g. min_x > max_x in profile-based observables).
  • Cylindrical observable classes have an extra transform_params argument to change the orientation of the cylindrical coordinates systems and control the origin of the phi angle (#4152).
  • Incompatible thermostat/integrator combinations raise an exception (#3880).
  • The system.cuda_init_handle.list_devices() feature is now a function, and the system.cuda_init_handle.list_devices_properties() function disabled in 4.0.0 was restored (#4095).
  • CUDA errors now halt the flow of the program by throwing a Python exception with a clear error message (#4095).
  • Parameter particle_scales of coupling-based fields PotentialField and ForceField now takes a dict object instead of a list of tuples (#4121).
  • The System class no longer has a globals member (#4276). Global variables are still accessible from other members of the System class.
  • Methods from the cluster analysis class Cluster no longer returns False when a string passed to call_method() doesn’t match the name of a core method; instead None is returned (#4234).
  • Methods from the cluster analysis class ClusterStructure, integrator classes and interaction classes no longer returns True when the corresponding core method doesn’t return a value; instead None is returned (#4234, #4516).
  • Several parameters of the ICC class are no longer optional: epsilons, normals, areas, sigmas (#4162).
  • The electrostatic actors charge neutrality check tolerance can be modified via actor.charge_neutrality_tolerance; this is mostly relevant to actors coupled to ICC, whose induced charges can have values spanning several orders of magnitude (#4506).
  • Electrostatic and magnetostatic methods that support tuning now have a timings argument to control the number of integration loops to run during tuning (#4276).
  • The Drude helpers (global variables and free functions) have been gathered into a checkpointable class DrudeHelpers, which now relies on particle handles instead of particle ids (#4353).
  • ScaFaCoS integration now supports activating an electrostatics ScaFaCoS actor at the same time as a magnetostatics ScaFaCoS actor (#4036).
  • The list of actors can no longer end up in an invalid state: updating an electrostatic or magnetostatic actor with an invalid parameter now automatically restores the original parameter; inserting an invalid electrostatic or magnetostatic actor in the list of actors now automatically removes the actor from the list of active actors, even when the exception happens during tuning (#4506).
  • The LBBoundaries slip velocity check was lowered to Mach 0.35, or 0.2 in LB units (#4376).
  • The OpenGL visualizer allows changing the radius of LB velocity arrows, documents all LB-related keyword arguments, and no longer suffers from a division-by-zero error that used to trigger a runtime warning for fluid inside boundaries (#4376).
  • The Electrokinetics class got an optional ext_force_density parameter for consistency with other LB implementations (#4203).
  • MDAnalysis integration now checks if the MDAnalysis package version is supported (#4386).

Removed functionality

  • The ENGINE shear torque calculation feature deprecated in 4.1.1 was removed (#3277).
  • The MEMBRANE_COLLISION and OifOutDirection features were removed (#3418).
  • The AFFINITY feature was removed (#3225).
  • The unused and untested UMBRELLA feature was removed (#4032, #4079).
  • The unused and untested VIRTUAL_SITES_COM feature was removed (#3250).
  • The unused and untested EK_DOUBLE_PREC feature was removed (#4192).
  • The unused and untested MD metadynamics feature was removed (#3563).
  • The unused and untested Stomatocyte shape was removed (#3730).
  • The PdbParser feature deprecated in 4.1.1 was removed (#3257).
  • The incorrectly implemented and untested HarmonicDumbbellBond interaction was removed (#3974, #4079).
  • The layered cell system was removed (#3512).
  • The unused Wang-Landau reaction ensemble algorithm was removed (#4288).
  • The reaction ensemble tutorial deprecated in 4.1.1 was removed (#3256).
  • The per-particle temperature feature was removed (#4057).
  • The Current observable was removed in favor of the FluxDensityProfile observable (#3973).
  • The incorrectly implemented analysis function cylindrical_average was removed in favor of the CylindricalDensityProfile observable (#3470).
  • The minimize_energy member of the System class was removed (#3390, #3891). The steepest descent algorithm is now a regular integrator that is set up via the system.integrator.set_steepest_descent() method.
  • The MMM2D electrostatics feature was removed (#3340). Electrostatics in slab geometries can still be achieved by ELC, with significantly better performance.
  • The dipolar direct sum with replica method is now disabled on periodic systems with zero replica, as it does not apply minimum image convention (#4061).
  • The analysis module min_dist2() function was removed and the dist_to() function was merged into system.distance_vec() (#3586).
  • The analysis module nbhood() function slab search mode was removed (#4516) since it was incorrect (all ESPResSo versions were affected).
  • The number of cells for the link cell algorithm can no longer be constrained to a range of values (#3701).
  • The global Mersenne Twister RNG was removed (#3482). All thermostats are now Philox-based. Local Mersenne Twister RNGs are still used in the linear polymer position generator (now with proper warmup) and in the ReactionAlgorithm class.
  • It is no longer possible to checkpoint an ESPResSo system instance that contains Union shape-based constraints when the simulation is running with 2 or more MPI ranks. An error will be raised (#4287, #4510).
  • It is no longer possible to checkpoint an Electrokinetics instance (#4327).
  • The unmaintained lj-demo.py sample was removed (#4482).
  • The unmaintained mayaviLive visualizer was removed (#4515).

Improved testing

  • The C++ core of ESPResSo is covered by unit tests and integration tests at 98% coverage (#4426, #4479, #4489).
  • The structure factor code is tested against simple lattices (#4205).
  • The MMM1D GPU code is tested (#4064).
  • The reaction method core classes are unit tested (#4164).

Performance enhancements

  • The Particle memory footprint was reduced and the MPI serialization code was improved (#4414). The structure size is now 584 bytes instead of 640 bytes on maxset configuration (10% reduction). All substructures in Particle are bitwise serializable and dynamic vectors are compact vectors. The performance gain is about 9% for a LJ liquid on both maxset and empty configurations, for both 1 000 and 10 000 particles per core.
  • Particle creation happens in constant time instead instead of monotonically increasing with the number of particles already in the system (#4493).
  • When only one MPI rank is used, the maximum cutoff of bonded interactions is ignored when initializing the cell properties, since the bond partners are always accessible on the same node, regardless of the cell size; if the system also doesn’t have short-range interactions, the short-range loop is skipped (#4452).
  • The ReactionAlgorithm::do_reaction() function used by reaction methods now caches the potential energy of the system and only updates it after a successful reaction trial move (#4374).
  • Reaction methods can delegate the particle neighbor search to the cell system when evaluating the exclusion range of inserted particles (#4401). This leads to better performance only on 2 or more MPI ranks.

Bug fixes

  • The transform_vector_cartesian_to_cylinder() now calculates the correct phi angle (#4094). The bug was present since ESPResSo 4.0.0 and affected observables CylindricalVelocityProfile, CylindricalFluxDensityProfile, CylindricalLBVelocityProfile, CylindricalLBVelocityProfileAtParticlePositions, CylindricalLBFluxDensityProfileAtParticlePositions.
  • Several memory leaks were fixed in the TabulatedBond interactions (#3961), electrostatics and magnetostatics tuning functions (#4069), lattice-Boltzmann code (#4108) and Barnes-Hut code (#4404).
  • The system.actors.clear() method was broken and would only remove half of the actors since 4.0.0. This is now fixed (#4037).
  • The ClusterStructure feature did not properly handle box periodicity since 4.0.0 and would under rare circumstances calculate a center of mass to be outside a fully periodic simulation, and would incorrectly fold coordinates in aperiodic systems. This is now fixed (#4363).
  • Adding a LB thermostat when any other thermostat was already active would silently fail since 4.0.0. This is now fixed (#4116).
  • Setting the NpT or steepest descent integrators with incorrect parameters no longer leaves the system in an undefined state (#4026).
  • The OpenGL visualizer had a tendency to slow down after pausing and resuming the simulation, or freezing when using the steepest descent integrator. This was due to a race condition between two threads that has been fixed (#4040).
  • The OpenGL visualizer no longer raises an exception when activating the LB_draw_boundaries option without any other LB_draw_* option (#4479).
  • The OpenGL visualizer now correctly updates bond information when the collision detection and bond breakage features are used (#4502).
  • It is no longer possible to accidentally set a non-cubic NpT integrator with P3M (#4165).
  • The NpT integrator used to work with P3MGPU even though it didn’t implement long-range energy calculation and therefore couldn’t contribute to the virial; now the long-range energy is calculated and added to the virial (#4026, #4506).
  • Illegal LB node access is now properly caught by exceptions (#3978).
  • EK node access no longer accepts floating-point values for node indices (#4228), and always requires exactly three integers (#4482).
  • Accessing the flux property of EK species no longer throws an error (#4106).
  • Accessing the boundary field of LB nodes from a LBFluid actor when LB_BOUNDARIES is not compiled in now returns 0 instead of a random integer (#4479).
  • The LB grid in the GPU implementation is now automatically resized when the simulation box size changes (#4191).
  • The LB code now throws an error when adding a LB boundary to the LBFluid actor when LB_BOUNDARIES is not compiled in, or to the LBFluidGPU actor when LB_BOUNDARIES_GPU is not compiled in (#4472).
  • The lattice-Boltzmann Python interface no longer ignores runtime errors, nor converts them to cryptic system errors (#4355).
  • The script interface no longer silently ignores runtime errors when converting Python objects to C++ data types (#4387, #4492).
  • The system now throws an error when a non-bonded interaction cutoff is too large for the local box size in MPI-parallel simulations; in older releases the error was queued and deferred to the integration loop (#4479).
  • The system now throws an error when a virtual site tracks a real particle too far away for the local box size in MPI-parallel simulations; in older releases the error was queued and deferred to the integration loop (#4479).
  • It is no longer possible for a virtual site to track itself (#4479).
  • It is no longer possible for a particle to exclude itself (#4493).
  • It is no longer possible to accidentally add the same bond twice on the same particles (#4058).
  • Fatal errors triggered by stale references in virtual sites, invalid particle ids and null quaternions have become runtime exceptions (#4479).
  • Virtual sites now contribute to the rotational kinetic energy of the system (#4198).
  • Particle creation no longer raises numpy.VisibleDeprecationWarning (#4493).
  • The EK feature now generates VTK files that are compliant with the VTK 2.0 standard (#4106).
  • The ELC and DLC actors now throw an error when a particle enters the gap region (#4051).
  • The ELC actor is now updated when the box size changes in the z-direction (#4231).
  • The DLC actor now raises an exception when tuning fails instead of causing a fatal error (#4238).
  • The MMM1D actor now raises an exception for incorrect periodicity or cell system instead of causing a segfault (#4064).
  • The DipolarP3M checkpointing mechanism was fixed (#3879).
  • The DipolarP3M method now recalculates the energy correction when the box length changes (#4506).
  • P3M-based actors now sanitize the user-provided alpha and accuracy parameters and no longer allow constraining the alpha parameter during tuning (alpha was always derived from the other parameters at the end of tuning) (#4118).
  • A buffer overflow in the DipolarP3M tuning function lead to random failures during tuning, this is now fixed (#3879).
  • A buffer overflow in the LB code could lead to incorrect results in grids of size 9x9x9 or larger with open boundaries, this is now fixed (#4078).
  • Providing incorrect parameters to the ScaFaCoS actors no longer cause ESPResSo to crash (#4068).
  • FENE, harmonic and quartic bonds now throw an error when the bond length is zero and the equilibrium bond length is non-zero, since the direction of the force cannot be determined (#4471).
  • Immutable parameters default_scale, particle_scales and gamma of coupling-based fields PotentialField, ForceField, FlowField and HomogeneousFlowField now throw an error when an attempt is made to change their value via the class setter, instead of silently ignoring the new value (#4121).
  • The CylindricalLBFluxDensityProfileAtParticlePositions observable now measures the correct quantity (#4152).
  • The Boost 1.74 bug was patched (#3978).
  • A bug involving an access out of bounds was fixed in the structure factor code (#4205).
  • A bug in the collision detection feature that lead to a harmless warning being printed to the terminal upon collision was fixed (#4484).
  • Calling collision_detection.set_params() with invalid arguments no longer leaves the collision detection feature in an indeterminate state; the previous state is automatically rolled back (#4484).
  • Setting the collision detection mode glue_to_surface or bind_at_point_of_collision when feature VIRTUAL_SITES_RELATIVE is not compiled in now generates the correct error message (#4484).
  • Passing a particle chain-based observable object (ParticleDistances, BondAngles, BondDihedrals, CosPersistenceAngles) that doesn’t have enough particle ids for the calculation (e.g. only 1 particle id when 2 are needed for the bond distance calculation) to a Correlator object no longer causes a memory overflow (#4255).
  • Calculating the energy of the system when an IBM object is present no longer terminates ESPResSo, instead a warning is issued (#4286).
  • The Sphere shape no longer returns NaN values in the distance vector for particles located exactly in its center (#4384).
  • Runtime errors raised when the maximal bonded interaction range becomes larger than the simulation box are no longer ignored when dihedral bonds are added to the list of interactions (#4383).
  • Runtime errors about incorrectly initialized electrostatic/magnetostatic methods are no longer silently ignored at integration start (#4383).
  • Runtime errors about incorrectly initialized GPU dipolar direct sum and Barnes-Hut are no longer silently ignored when the actors are instantiated (#4404).
  • A bug that could potentially lead to stale references in the script interface was fixed (#4476).
  • TabulatedNonBonded.is_active() now returns False instead of None when the interaction is inactive (#3586).

Under the hood changes

  • The Python code is now checked with Pylint to prevent the introduction of unused code and dangerous anti-patterns (#3293, #3203).
  • The CMakeLists.txt files are now formatted automatically with cmake-format (#3622).
  • The Python code and C++ code were checked with LGTM to detect and fix coding errors and anti-patterns (#3851, #3856, #4300).
  • Compiler warnings and diagnostics from GCC 11, 12, from Clang 10, 12, 13, 14 and from Intel 19.0.4 were addressed (#4084, #4426, #4510, #4526).
  • The Particle struct was moved to a dedicated header file Particle.hpp to improve separation of concerns in the core (#3251, #3164).
  • The Observable_stat structs were moved to a dedicated header file Observable_stat.hpp and decoupled from the pressure/energy/coulomb/dipolar frameworks (#3712) and made stateless (#3723).
  • Observables based on particle ids have been rewritten using particle traits to decouple the Particle struct from Observable classes (#3667).
  • The Python Integrator class was split into multiple classes, one for each integrator, with a structure similar to actor and interaction classes (#3390). This layout better reflects the structure of integrators in the core and will make it easier to include new integrators in the future. This change doesn’t break the API.
  • The ghost communication infrastructure was simplified (#3216, #3399).
  • The LB coupling for the regular decomposition scheme was rewritten (#4470).
  • Thermostats are now fully object-oriented in the core to reduce code duplication (#3438, #3444, #3461).
  • Bonded interactions are now fully object-oriented in the core to facilitate the development of new interactions (#4161).
  • Bonded interactions are now communicated between MPI processes automatically and transparently by the script interface (#4350).
  • The custom MpiCallbacks framework has been simplified and the callbacks made more homogeneous (#4383).
  • The custom MpiCallbacks framework is being progressively replaced by boost::mpi communication (#4506, #4511).
  • The local_particles global variable is no longer accessible directly (#3501).
  • The Python tests now use specialized assertions to generate more helpful error messages (#3419).
  • The tutorial tests were simplified using AST to parse Jupyter notebooks (#3408).
  • The CMake logic for tutorials has been simplified (#3408, #3486).
  • The Cython interface was thoroughly cleaned up from unused imports (#3496, #3510).
  • The ScriptInterface framework was rewritten (#3794).
  • The ScriptInterface framework is now the preferred way to implement new features. Existing features were converted to ScriptInterface objects: bonded interactions (#4350), bond breakage (#4464), collision detection (#4484), reaction methods (#4451), MPI-IO (#4455), H5MD (#4480), cell system (#4511), actors, scafacos, electrostatics and magnetostatics (#4506). The corresponding Cython files were converted to Python files.
  • It is now possible to extend the list of available specifications in the H5MD feature at the C++ level (#4480).
  • The duplicated functions between P3M and DipolarP3M were factored out (#3879).
  • Statistical tests are no longer executed in coverage and sanitizers builds (#3999).
  • The Utils::Mpi::gather_buffer() function was fixed (#4075). The bug didn’t affect ESPResSo.
  • Parameters can be passed to CTest at configuration time via the new CTEST_ARGS CMake option (#3862). This replaces the deprecated and non-portable ARGS Makefile variable expansion.
  • A superfluous and non-portable CMake target_compile_options() statement was removed (#3852).