Kinetic models for micro-scale transport phenomena and structure formation in complex fluids:

implementation on GPU-based parallel computing systems

Exploratory Research Project PN-II-ID-PCE-2011-3-0516

2011 - 2014

supported by CNCS - UEFISCSU




    Project Summary

The project is dedicated to the development and application of three-dimensional (3D) Lattice Boltzmann (LB) models for the study of complex fluids and their micro-scale behaviour. LB models allow the easy incorporation of interparticle forces as well as of boundary conditions and are appropriate for implementation on massively-parallel computers. Intensive simulations will be done on the IBM Blue Gene / P system recently inaugurated at the West University of Timisoara, as well as on a "desktop supercomputer" incorporating two Graphics Processing Units (GPU) to be set up at the host Institution. Simulations will be conducted in order to understand the effect of various conditions (fluid viscosity, temperature gradients and/or anisotropy induced by gravity or shear flow on the phase separation process and the scaling exponents that characterize the domain growth in single- or multi-component fluid systems. The morphology of the 3D structures formed durinig the phase separation, as well as their dynamics, will be investigated using the Minkowski functionals. Specific problems in microfluidics will be also investigated through 3D Lattice Boltzmann simulations: rarefaction effects in long microchannels, temperature dip in force-driven flow and thermal creep. The formation of drops or bubbles introduced in the co-flowing phase via a capillary tube or a T-junction will be another process to be analyzed through LB simulations in three dimensions.


    Objectives
  1. Development of high order 3D LB models for phase-separating fluids, as well as of appropriate numerical schemes using the PETSc library for parallel computing on multi-core CPU systems.
     
  2. Build up of a GPU-based computing system and implementation of LB models on the many-core system using the CUDA library.
     
  3. Investigation of the phase transition dynamics and morphology of 3D systems using the GPU parallel computing system.
     
  4. Application of 3D Lattice Boltzmann models to specific problems in microfluidics.


    Publications
  1. B.Piaud, S. Blanco, R.Fournier, V.E.Ambrus, V.Sofonea

    Gauss quadratures - the keystone of Lattice Boltzmann models

    submitted to International Journal of Modern Physics (2012).



    Conference presentations
  1. B.Piaud, S. Blanco, R.Fournier, V.E.Ambrus, V.Sofonea

    Gauss quadratures - the keystone of Lattice Boltzmann models

    21st International Conference on Discrete Simulation of Fluid Dynamics

    July 23 - 27, 2012, Bangalore (India)
     

  2. V.Sofonea, V.E.Ambrus

    Diffuse Reflection Boundary Conditions and Lattice Boltzmann Models for Microfluidics

    CECAM Workshop Fluid-Structure Interactions in Soft-Matter Systems: From the Mesoscale to the Macroscale

    November 26 - 30, 2012, Prato (Italy)



    Computing resources
  1. GPU-based computing system SUPERMICRO with 2 Quad-core CPUs and 48 GB RAM memory (photos below). The system has 4 GPUs nVIDIA Tesla Fermi M2090. Each GPU unit has 512 cores and 6GB internal memory.
     
  2. Parallel computing cluster : 1 MacMini server, 2 MacPro (each with 2 Quad-core CPUs), 2 iMac (quad-core CPU) and 4 MacMini (dual core CPU).
     
  3. Access to the IBM Blue Gene / P system at the West University of Timisoara.