ARCHEReCSEJuneMinutes

Minutes meeting on 11-12 June 2015

Agenda: http://pauli.chem.ucl.ac.uk/trac/hemelb/wiki/ARCHEReCSEJune

Summary of actions

MAYEUL: Finish task 5 (particle iolets).
MAYEUL, GARY: We will implement cell growth in the virtual inlet. Cell growth with walls. Code will figure out where to place the cells before growing them based on knowledge of cylindrical domain.
MIGUEL: VMTK flow extensions.
GARY: Task 9: Cell output via HDF5 (or derived) and postprocessing scripts to generate Paraview vis.
ALL: Get started with literature review for first paper and follow-up grant. See Timm's e-mail on 12 June. Deadline 26 June.

= Review time spent =

We have worked 59% of the total hours. We believe 85% will be achieved by end of August. However, tasks 7 and 8 might be optimistically scheduled.

= Review progress to date =

Task 5

At the moment we have a:

A cell distribution. Currently columns of cells, but could be another flow simulation as in Janoscheck2013.
A conveyor belt that moves particles along (based on the previous cell distribution).
The conveyor belt handles cells to a cell controller that places them in the drop zone defined within the flow extension.
Cells automatically fade in/out based on their own knowledge of whether they are in flow extension.

In order to finish the task, we need:

A mechanism for migrating cells from one domain (conveyor belt) to another (drop zone). This will be done as part of Task 6.

Task 6/7

No way of knowing comms needs at the beginning of the simulation.

At each time step:

We have code to compute communication needs, but don't know whether can be integrated into coalesced comms.
There are methods for issuing the MPI comms, which are unit tested, but not calling MPI yet.

The parallel algorithm to be implemented is:

Interpolate (IBM) - computation - done.
Collect nodes (collect velocities corresponding to nodes that are physically on a different subdomain) - comms - not done.
Update position - computation - done.
Migrate cells - comms - partially done.
Compute forces - computation - done.
Distribute nodes (send node index, forces, etc, similarly to collect nodes) - partially done.
Compute particle-particle interaction - computation - done.
Spread forces (IBM) - comms - done.
Solve LBM - computation - done.
Iterate

Task 8

Nothing done beyond the potential overlaps with other tasks.

Preliminary simulations

Node location written out in lattice units.
Moduli are currently set to 0 in large_cylinder_rbc.xml. Instead use:
Don't expose volume and surface moduli to the user, set them to 1 in lattice units.
k_strain = 5e-6 N/m
k_bending = 2e-19 Nm
k_dilation = 0.5 N/m
We see abnormal deformation. Possible causes:
Velocity interpolation.
Position update.
Test both the previous with a constant velocity field with unit tests.

Planning ahead

Planning for 2.5 months until end of August.

Gary and Mayeul to work full time (combined) until end of August. That equates to 2 months worth of work if we take into account holidays.

For the ICCB poster we will focus on a small network of capillaries and run a simulation with a realistic hematocrit of 10-15%.

What do we need for the poster at ICCB:

Finish task 5 (particle iolets). Mayeul.
We will implement cell growth in the virtual inlet. Cell growth with walls. Code will figure out where to place the cells before growing them based on knowledge of cylindrical domain. !Mayeul/Gary.
VMTK flow extensions. Miguel
Task 9: Cell output via HDF5 (or derived) and postprocessing scripts to generate Paraview vis. Gary.

In terms of parallelisation, CellController implements IteratedAction interface. As a first approximation we will let it issue its own MPI comms (via HemeLB's wrapper class) instead of interacting with the coalesced communication.

Simulation timestep has multiple phases. Each phases has multiple steps. Each phase implements a communication cycle (a single synchronisation point).

The parallel algorithm to be implemented is:

VelocityPhase: Interpolate (IBM)
VelocityPhase: Collect nodes (collect velocities corresponding to nodes that are physically on a different subdomain)
VelocityPhase: Update position
ForceAndOwnershipPhase: Migrate cells
ForceAndOwnershipPhase: Compute forces
ForceAndOwnershipPhase: Distribute nodes (send node index, forces, etc, similarly to collect nodes)
ForceAndOwnershipPhase: Compute particle-particle interaction
ForceAndOwnershipPhase: Spread forces (IBM)
!LBPhase: Solve LBM
Iterate

Publication plan

Journals: JOCS, CPC, PLOS Comp Bio, Frontiers in Physiology, SIAM Scientific Computing, eLife.

What does it enable:

Microfluidics: applications that require sparse geometries. Cell separation.
Droplet microfluidics. Droplet flow control, sorting. Biopharma. Single cells within a droplet.
Hemorheology microvasculature. Plasma separation. Understanding design principles of branching patterns.

What's technically novel:

IBM with sparse geometries. Its parallelisation.
Applications that require big systems.

Timeline:

!Miguel/Timm write introduction in August. Continue with methods until after ICCB.
While doing the previous we should come up with concrete ideas for simulation studies (hypothesis).

Follow-up grants

Options:

Look for relevant H2020 calls and go for big application including multiple partners.
Looks more like BBSRC remit than EPSRC.

Potential projects:

Oxygen delivery in the retina. Diabetic retinopathy.
Drug delivery in tumours.
Distribution in cells in general networks. Design principles of cardiovascular system.
Hematocrit going down in microvasculature means lower viscosity and lower resistance to flow. Does this help with reducing cardiac work?
Implications of Murray's law to cell distribution at branches.
Oxygen delivery wants increased contact area. RBC need to be soft. In large vessels cells don't get close to the walls. How did softness and small vessels codeveloped?
Axel Pries, Tim Secomb. Studies on oxygen delivery, hematocrit, etc. Check out latest 5 papers and look for gaps in the literature.