Fluid-Particle Simulation With CFDEM®coupling

It's the CFD-DEM platform that runs Aspherix® (DEM) and the fluid solver side by side, exchanging voidfraction, momentum, mass, and energy every coupling step, a full 4-way coupling. Beyond the coupling mechanism itself, CFDEM®coupling supplies its own library of dedicated solvers and physics models, incompressible, compressible, reactive, Volume-of-Fluid, and Euler-Euler-DEM, resolved CFD-DEM (immersed boundary method) among them. Domain decomposition and parallelization for the particle side and the fluid side are independent, which pays off when the particle phase and fluid phase occupy very different parts of the domain (shafts, furnaces, packed beds).

No. CFDEM®coupling supports unresolved (particle smaller than the cell), resolved (particle larger than the cell, via Immersed Boundary), and dynamically hybrid representations, where individual particles switch between the two at runtime as the local mesh resolution changes.

The full shape library carries through into CFD-DEM: spheres, multisphere clumps, convex and concave triangulated (faceted) shapes, superquadrics, boxes/cylinders/ellipsoids, fibers, and bonded-particle assemblies. Coupled voidfraction mapping and Immersed Boundary methods have dedicated formulations for convex and superquadric shapes specifically, and melting is supported across all of them (sphere, multisphere, faceted, superquadric, and bonded). 

Both, via dedicated solver families: incompressible and compressible fluid-particle flow, a Volume-of-Fluid solver for resolved gas–liquid interfaces with melting/solidification, a Euler-Euler-DEM solver for interpenetrating fluid phases (e.g. gas-stirred liquids, bubble columns), and a reactive solver built on OpenFOAM®'s chemistry/combustion framework. Systems with up to five phases and non-spherical particles have been run.

The Volume-of-Fluid solver family models particles together with multiple fluid phases including particle–fluid phase change, for example solid charge material floating at a liquid interface, heating up, and melting into the bath while losing mass and size.

Yes, radiative heat transfer for fluid-granular systems is available and is aimed specifically at high-temperature applications.  

Millions of particles are natively achievable without any special treatment. Beyond that, coarse-graining is the standard approach for most industrial applications: several physical particles are lumped into one numerical parcel, cutting the number of numerical objects while preserving contact-based physics rather than replacing it with a statistical closure. Scaling laws are applied automatically and consistently to insertion rates, contact stiffness, drag, and fluid–particle exchange, so the correct bulk behavior is fully maintained. With coarse-graining, applications involving billions of physical particles become tractable, with reactive industrial cases already run at up to ~10⁷ numerical parcels and the largest benchmarked non-reactive coupled case at ~10⁶ particles matched to ~10⁶ CFD cells.

Yes for non-reactive single-fluid coupled cases both the CFD and DEM sides can run on GPU (Linux only), showing roughly 2× speed-up over CPU at comparable compute power. Reactive flows are CPU-only at present.

As a rule of thumb, 32+ CPU cores. A reference reactive case (iron ore reduction, ~0.4–11 million parcels depending on coarse-graining) ran on 62 AMD EPYC cores at wall-clock speeds from roughly 1.6 to 47 hours per second of simulated time, depending on coarse-graining level.

No. Standard MPI 3 compliance is the only requirement, so any modern CPU works, and near-linear scaling has been demonstrated to 1,000 cores with normal Slurm/PBS/LSF job submission.