Skip to content

4.1.0

Compare
Choose a tag to compare
@jngrad jngrad released this 01 Oct 12:03
· 459 commits to 4.1 since this release
3ffa392

This feature release adds new functionality to ESPResSo. An additional focus of this release is quality assurance and modernization. The testing of ESPResSo's functionality has been extended considerably. Also, sample and tutorial scripts are now automatically tested. Moreover, a large effort was put into modernizing the C++ simulation core. Work has been done, e.g., on particle sorting, the MPI communication infrastructure, and the lattice-Boltzmann implementations. Electrostatic and magnetostatic methods now have a clear and common interface. These changes will facilitate future extensions of ESPResSo and make the code more understandable to new developers.

Changed requirements

  • Python 2 support has been dropped. ESPResSo now requires Python 3.
  • ESPResSo now needs a C++14-capable compiler, such as GCC 4.9 and later or Clang 4 and later.
  • It is discouraged to use ESPResSo with Boost versions below 1.67.

Added functionality and documentation

  • The distance between a shape (such as sphere) and a position can now be queried via shape.calc_distance().
  • The lattice nodes of a lattice-Boltzmann fluid can now be iterated using LBFluid.nodes().
  • A tutorial on magnetic fluids has been added.
  • The stress created by the dissipative particle dynamics interaction (DPD) can now be obtained via the DPDStress observable.
  • The stress of a lattice-Boltzmann fluid can now be obtained via the LBFluidStress observable.
  • A torus shape has been added.
  • Two new accumulators for observables have been added: MeanVarianceCalculator and TimeSeries.
  • An ElectricPlaneWave constraint was added.
  • Experimental support for AMD GPUs via HIP. The future of this feature is unclear. Please do not base hardware buying decisions on its presence.
  • Visualization of slit pores in the OpenGL visualizer.
  • A Weeks-Chandler-Anderson short-range potential has been added (WCA).
  • The external force density applied to a lattice-Boltzmann fluid can now be changed during the simulation.
  • Sanity checks for Mach limits and unequal MD and lattice-Boltzmann time steps have been added.
  • The system.cell_system.tune_skin() method now has a keyword argument adjust_max_skin. If set to True, the maximum skin to be tested will be reduced such that it is compatible with the local box size.
  • For the CPU lattice-Boltzmann implementation, the limit on the Verlet list skin (<0.5 agrid) has been lifted.
  • A new observable CosPersistenceAngles has been added for the bond angles of a polymer (needed, e.g., for determining the persistence length).

Feature configuration at compile time

  • The number of features which need to be defined at compile time in myconfig.hpp has been reduced. Features without performance impact are now always present. These are:
    • PARTIAL_PERIODIC
    • LB
    • LB_GPU (for builds with CUDA)
    • IMMERSED_BOUNDARY
    • BOND_ANGLE
  • For most compilers, it is checked that only known features are declared in myconfig.hpp.
  • The default feature configuration applied when no myconfig.hpp is present has been extended significantly. In particular, all tutorials can now be run with the default feature configuration. For production simulations, it is still recommended to use a custom myconfig.hpp containing only necessary features. This is true in particular, if particle rotation is not needed.
  • Features that do not have automated tests now require EXPERIMENTAL_FEATURES to be defined in myconfig.hpp.

Interface changes

Changed and removed functionality

  • The remove_total_momentum() method for lattice-Boltzmann fluids has been removed. The overall velocity of a fluid can be changed using the lb_fluid.nodes() iterator.
  • The CATALYTIC_REACTIONS feature has been removed.
  • The method for creating a polymer has been replaced. espressomd.polymer.positions() can now be used to obtain particle positions for one or more polymer chains.
  • Checkpointing has been added for the electrokinetics method.
  • The global random number seed has been partly replaced by method-specific ones. These are specified when activating the relevant feature such as the Langevin, DPD and lattice-Boltzmann thermostats via a seed keyword argument.
  • The random number generator has been switched to Philox for most algorithms requiring random numbers.
  • Limitations on the exclusion radius have been relaxed in the reaction ensemble method.
  • A new observable CosPersistenceAngles has been added for the bond angles of a polymer (needed, e.g., for determining the persistence length).
  • ELC has been disabled for non-neutral systems with constant potential.
  • The calculation of the linear particle momentum included the forces of the last time step. The function system.analysis.linear_momentum() now returns the sum of the product of mass and velocity of all particles, if no lattice-Boltzmann fluid is coupled.

Performance enhancements

  • Speedup in the short-range force calculation in situations where the short-range cutoff varies strongly for different pairs of particles, e.g., in a bidisperse fluid.
  • Speedup in particle resorting triggered when particles have moved by more than a skin.
  • Significantly faster back-transfer of particle forces from GPU-based methods such as the GPU implementations of lattice-Boltzmann and P3M.

Bug fixes

  • Lattice-Boltzmann boundaries, constraints and auto_update_accumulators are now included in checkpointing. (#2915)
  • Collision detection is now checkpointed. (#2342)
  • The rhomboid shape was fixed. (#2756)
  • Deadlocks on certain GPUs have been resolved for the dipolar Barnes-Hut method. (#2719)
  • The visualization of dihedrals has been fixed. (#2677)
  • The ENGINE implementation for CPU LB has been fixed. (#3025)
  • The external force density in lattice-Boltzmann fluids is no longer ignored in the first integration step after setting the force density. (#3144)
  • The positions of virtual sites and the charges of ICC particles are now updated before observable calculation. (#3128)
  • The forces and torques for the Gay-Berne potential have been corrected. (#3091)
  • Remove undocumented behaviour in the case of using a cylindrical sampling area in the reaction ensemble, constant pH ensemble, Wang-Landau ensemble, Widom-Insertion method. (#3174)
  • The ELC tuning error calculation has been rearranged to produce correct results for higher accuracies. (#3123)

New tutorials

  • Tutorials for simulating ferrofluids and for using the constant-pH method have been added.

Under the hood changes

  • Automated testing has been enhanced. It now also includes samples and tutorials. The overall test coverage for the simulation core has increased by ~12% since ESPResSo 4.0.2.
  • The CPU LB and LB-particle coupling have been refactored.
  • Particle resorting has been simplified and sped-up.
  • The MPI callback mechanism has been simplified. Furthermore, reduction operations such as summing values from all MPI ranks can now be performed.
  • Nearly all manual memory management and C-style arrays have been removed.
  • The rotation-related code has been simplified.
  • Long-range electrostatic and magnetostatic methods now have a common interface.
  • The kernels for short-range and bonded interactions have been simplified.
  • The CMake build system has been refactored and dependencies between different parts of the code have been made clear.
  • Python code formatting: the autopep8 version now matches the one in Ubuntu 18.04 (autopep8 v1.3.4 with pycodestyle v2.3.1).