search

hemelb-codes / hemelb

A high performance parallel lattice-Boltzmann code for large scale fluid flow in complex geometries
GNU Lesser General Public License v3.0
35 stars 11 forks source link

Implement rigid IBM algorithm #736

Open mobernabeu opened 4 years ago

mobernabeu commented 4 years ago

Following from #569 and #684, the current algorithmic ordering in HemeLB is:

  1. compute RBC forces
  2. spread forces to fluid grid (new \vec{F}(\vec{x}, t))
  3. interpolate velocities at Lagrangian mesh (using \vec{u}^*(\vec{x}, t-1) and \vec{F}(\vec{x}, t), see #684)
  4. update positions of Lagrangian mesh
  5. LBM collision
  6. LBM propagation
  7. swap current and new (collided + propagated) populations (this is an important implementation detail: https://github.com/UCL-CCS/hemelb-dev/issues/684#issuecomment-299882110)

Note: \vec{u}^*(\vec{x}, t-1)=1/\rho \Sum_i f_i(\vec{x}, t-1) \vec{c_i} is the collided, pre-streamed velocity in the previous timestep.

The current RBC development branch is red_blood_cells/develop. The relevant files are Code/redblood/CellController.h for the first four steps and Code/lb/lb.hpp for the last three steps. Both are subclasses of the IteratedAction infrastructure, which will require some study to understand how things are orchestratred together. Documentation is sparse so I'm happy to give more details. No harm in adding some more documentation as we go along.

There are some known issues with unit tests not passing in #693.

otbrown commented 4 years ago

Hi all (in particular @timmkrueger),

Apologies for the lengthy comment, but I thought it might be useful to present my notes here. I've been looking through HemeLB, and taking note of parts of the code which I may (or may not) need to modify. Having identified where the steps noted above by Miguel are implemented, I've now begun looking at how Emmanouil's algorithm would be implemented.

As you'll see I've not got terribly far, but I think it would be useful to have someone who knows HemeLB well confirm or deny my suspicions about what's already implemented. It would be also useful to discuss with Emmanouil if he thinks existing code correlates to his algorithmic steps. Is he on here? I can send this separately to him if not.

All comments and questions welcomed!

Cheers, Oliver

Current Ordering

(Assuming HEMELB_USE_KRUEGER_ORDERING, using namespace hemelb::redblood) Algorithmic Step Function(s) Call Definition
1: Compute RBC forces CellArmy<TRAITS>::Cell2FluidInteractions > parallel::SpreadForces::ComputeForces CellController.h:38 > CellArmy.h:291 CellArmy.h:284 > SpreadForces.cc:40
2: Spread forces to fluid grid CellArmy<TRAITS>::Cell2FluidInteractions > parallel::SpreadForces::SpreadLocalForces, parallel::SpreadForces::SpreadNonLocalForces CellController.h:38 > CellArmy.h:293,297 CellArmy.h:284 > SpreadForces.h:108,123
3: Interpolate velocities at Lagrangian mesh CellArmy<TRAITS>::Fluid2CellInteractions > parallel::IntegrateVelocities::ComputeLocalVelocitiesAndUpdatePositions > velocitiesOnMesh > interpolateVelocity CellController.h:41 > CellArmy.h:243 > IntegrateVelocities.h:89 > GridAndCell.h:40 CellArmy.h:215 > IntegrateVelocities.h:79 > GridAndCell.h:31 > VelocityInterpolation.h:131
4: Update positions of Lagrangian mesh CellArmy<TRAITS>::Fluid2CellInteractions > parallel::IntegrateVelocities::ComputeLocalVelocitiesAndUpdatePositions > CellBase::operator+= CellController.h:41 > CellArmy.h:243 > IntegrateVelocities.h:90 CellArmy.h:215 > IntegrateVelocities.h:79 > CellBase.cc:132
5: LBM collision* LBM<TRAITS>::PreSend, LBM<TRAITS>::PreReceive > LBM<TRAITS>::StreamAndCollide During phase 1 > lb.hpp:many lb.hpp:217, lb.hpp:255 > lb.h:132
6: LBM propagation * LBM<TRAITS>::PreSend, LBM<TRAITS>::PreReceive > LBM<TRAITS>::StreamAndCollide During phase 1 > lb.hpp:many lb.hpp:217, lb.hpp:255 > lb.h:132
7: Swap current and new populations LBM<TRAITS>::EndIteration > geometry::LatticeData::SwapOldAndNew End of phase 1 > lb.hpp:357 lb.hpp:350 > LatticeData.h:67

* Does StreamAndCollide, so not sure it makes sense to separate these two steps. Actual StreamAndCollide operation is dependent on location (near a wall, mid-fluid, etc) and provided Streamer and Collision, see lb/streamers and lb/collisions directories.

Special mention: TODO at lb.hpp:308.

Proposed ordering

Algorithmic Step Implemented in HemeLB?
1. Total force computation Yes, if same as Compute RBC Forces above.
2. CoM velocity prediction Possibly.
3. Particle Position Yes, assuming we can use existing particle position update functions. (ComputeLocalVelocitiesAndUpdatePositions)
4. Velocity computation Yes, assuming we can use existing velocity update functions. (ComputeLocalVelocitiesAndUpdatePositions)
5. CLBM Collision step, yes, although may need to separate from streaming step.
6. Velocity Interpolation Yes, interpolateVelocity.
7. Lagrangian velocity correction Possibly, part of IntegrateVelocities?
8. Lagrangian force computation Possibly.
9. Eulerian velocity correction Possibly.
10. Eulerian force computation Possibly.
11. Iteration No, this is trivial if all iterated calculations are local, as we can confine new steps to one step (PreReceive). If communication is required could be more challenging.